Datasets:

Abirate
/

all_write_ups

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 3.85k rows

Subsets and Splits

Title of Competition stringclasses 343 values	Title of Writeup stringlengths 6 139	User stringclasses 1 value	Writeup stringlengths 419 80.4k
Global Wheat Detection	[34th Place Solution using single EfficientDet + Pseudo Labeling]	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Global Wheat Detection <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Sure! I've updated the description with links to public notebooks.</p> <p><a href="https://www.kaggle.com/brandonnova" target="_blank">@brandonnova</a> Thanks!</p> <p><a href="https://www.kaggle.com/brendan45774" target="_blank">@brendan45774</a> Thanks!</p> <p><a href="https://www.kaggle.com/piantic" target="_blank">@piantic</a> Thanks!</p> <p>Yes it was single model PL. The baseline was really good and I didn't change it much. I just trained on entire dataset as I wanted to use it in PL. The main change I did was in PL training though. I used same augmentations which I used in training (changing augmentations reduced my score). And even no data cleaning! So I think model might be underfitting the training data but generalized comparatively well. P.S This was my first object detection competition and this was completely unexpected!</p> <p>First of all Congratulation to all, especially to the prize winner and gold medalists and thanks to the organizers for this interesting competition!</p> <p>So here is our solution to this competition (Very very basic):</p> <h2>Training</h2> <ul> <li>Single EfficientDetD5 (Thanks to <a href="https://www.kaggle.com/shonenkov" target="_blank">@shonenkov</a> for his great notebook)</li> <li>40 epochs using full dataset (No CV)</li> <li>Image size: 512 * 512</li> <li>Augmentations: Random Crop, Random Cutout, Hue Saturation Value, Brightness Random Brightness Contrast, Random Horizontal and Vertical flips and random Cutmix.</li> <li>Training notebook: <a href="https://www.kaggle.com/kaushal2896/gwd-efficientdet-training-on-entire-dataset" target="_blank">https://www.kaggle.com/kaushal2896/gwd-efficientdet-training-on-entire-dataset</a></li> </ul> <h2>Pseudo Labeling</h2> <ul> <li>Pseudo labeled the above model with 10 epochs (Thanks to <a href="https://www.kaggle.com/nvnnghia" target="_blank">@nvnnghia</a> for his work)</li> <li>Used the same augmentations even while pseudo labeling: Random Crop, Random Cutout, Hue Saturation Value, Brightness Random Brightness Contrast, Random Horizontal and Vertical flips and random Cutmix.</li> <li>Same image size: 512 * 512.</li> <li>Used both TTA and WBF for final predictions.</li> <li>Pseudo Labeling notebook: <a href="https://www.kaggle.com/kaushal2896/efficientdet-pseudo-labeling-wbf-tta" target="_blank">https://www.kaggle.com/kaushal2896/efficientdet-pseudo-labeling-wbf-tta</a></li> </ul> <h2>What didn't work for us?</h2> <ul> <li>Custom augmentations like mixup.</li> <li>Training on larger image sizes without Random crop augmentation.</li> <li>Same solution with EfficientDetD6.</li> <li>Training on higher number of epochs.</li> <li>FasterRCNN with ResNet51 and ResNet101.</li> </ul> <p>Here are different approaches I've tried during this competition: <a href="https://github.com/Kaushal28/Global-Wheat-Detection" target="_blank">https://github.com/Kaushal28/Global-Wheat-Detection</a></p>
Global Wheat Detection	[42nd place solution]	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Global Wheat Detection <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thanks for your suggestion <a href="/aidynub">@aidynub</a> </p> <p>No, I didn't</p> <p>Here is one of my experiments nearly 2 months ago using fasterRCNN<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2545270%2F693038ebe35bee667f43fdb798263338%2FScreenshot%20from%202020-08-07%2016-10-52.png?generation=1596784304454586&alt=media" alt=""><br> after this experiment, SPIKE dataset is my best friend.</p> <p>by trial-error. FasterRCNN gives a better result than D5</p> <p>Congratulation to all, especially to the prize winner and gold medalists and thanks to the organizers for this interesting competition.<br> So here are some essential points of our solutions (I will update more detail later)<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2545270%2F4c4ce442608f2f655f7182babf7a7bca%2FUntitled%20Diagram%20(2).png?generation=1596588070622865&alt=media" alt=""><br> <strong>1. Validation strategy</strong><br> I use a fixed validation set from the beginning of the competition (663 images). The validation data is from usask_1 and a part of other sources.<br> <strong>2. Augmentation</strong></p> <p>2.1. My custom augmentation: <br> - crop and pad: I realize that the boxes at the border after cropping is not good. Then I remove all the boxes at the border and remove the image content inside the boxes too.<br> - crop and resize<br> - resize and pad<br> - <a href="https://www.cs.tau.ac.il/~turkel/imagepapers/ColorTransfer.pdf" target="_blank">Color Transfer between Images</a><br> - mixup: Instead of mixup 2 images in training set, I use 1 image in training set and 1 image without wheat head. <br> - Rotation: random rotate images and use pseudo label as ground truth.</p> <p>2.2. Albumentation:<br> - A.HorizontalFlip(p=0.5),<br> - A.VerticalFlip(p=0.5),<br> - A.RandomRotate90(p=0.5), <br> I don't use mosaic augmentation because it does not want to cooperate with my custom augmentation.<br> <strong>3. SPIKE dataset:</strong> I use WBF on pseudo label and provided ground truth label.<br> <strong>4. Results</strong><br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2545270%2F9312433df689b4ebe397f0674b067b9b%2FUntitled.png?generation=1596632145633431&alt=media" alt=""></p> <p>P/s: Is there a chance for us to move into gold zone after verifying solution? </p> <p>Thanks <a href="/piantic">@piantic</a> - I retrain fasterRCNN for 7 epochs. I use time tracking on my notebook, so my notebook will stop the training process whenever my notebook meets the time limit (I set it to 5 hours) - I kept a fixed validation set at the beginning of the competition (663 images). The validation data is from usask_1 and a part of other sources.</p> <p>resnet152</p> <p>Thank <a href="/matthewmasters">@matthewmasters</a> Thanks for your huge contribution in this competition. Yes I use an ensemble of EffDet models for the pseudolabel prediction and FasterRCNN for retraining. </p> <p>detail updated!</p> <ul> <li>FasterRCNN 0.7582, </li> <li>D0: 0.7605</li> <li>D5 (512x512): 7584</li> <li>D5 (768x768): I didn't check LB score. But I know it is good because of high CV score.</li> </ul> <p>I will update more detail later. </p> <p>I ignore images without bboxes. Just remember to add an empty string to submission file.</p> <p>TTA only</p>
Global Wheat Detection	[80th place solution😂] Source-separated Psuedo Labeling	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Global Wheat Detection <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thanks! The idea is simple but is quite complicated to implement.. ;)</p> <p>Thank you!</p> <p>Thank you!</p> <p>Big thanks to the whole team ;)</p> <p>Thanks for the warm words! Let's see after the YOLO subs are removed..</p> <p>First of all, thanks for the host for holding this competition! </p> <p>And great thanks to our team members, <a href="https://www.kaggle.com/datahobbit" target="_blank">@datahobbit</a>, <a href="https://www.kaggle.com/ratthachat" target="_blank">@ratthachat</a>, <a href="https://www.kaggle.com/yashchoudhary" target="_blank">@yashchoudhary</a>, <a href="https://www.kaggle.com/nitindatta" target="_blank">@nitindatta</a> and it was a pleasure teaming up. Everyone played a great part in the solutions which <a href="https://www.kaggle.com/ratthachat" target="_blank">@ratthachat</a> elaborates! It was fun having discussions and everybody was a great analyst/coder. We did a great job and made it into the TOP-2% ;)</p> <p>This is our solution of our team Wheat Fields!</p> <p>(I will come back to update with the details.)</p> <h2>Models</h2> <ul> <li>Model: EffdetD5 2 folds ensemble </li> </ul> <p>We evaluated our CV based on data source separated CV with original weightings.</p> <h2>Source separated psuedo labeling</h2> <h3>Idea</h3> <p>We were scared that psuedo labeling with the entire test set prior will overfit to the UTokyo data which should be 70% of the test data. With this source-split, we can train model which overfit to UTokyo and other model to the rest, maximizing the PB.</p> <h3>Step.1</h3> <p>Predict on test set using Best CV Model and split the test set to two: TestSetB :UToyko which image has <code>number of wheats <39</code> and TestSetC: <code>number of wheats>=39</code>. We save the labels for psuedo label as well</p> <h3>Step2:</h3> <p>train two models --> Model-B train with pseudo [TestSetB, traindata] and Model-C train with pseudo [TestSetC,traindata].</p> <h3>Inference:</h3> <p>Predict TestSetB with Model-B and TestSetC with Model-C.</p> <h1>Codes!</h1> <p>I uploaded my training codes here.<br> <a href="https://github.com/kentaroy47/kaggle-wheat-arutema47" target="_blank">https://github.com/kentaroy47/kaggle-wheat-arutema47</a></p>
Global Wheat Detection	[From 447 to 135th place Solution] EfficientDet Ensemble	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Global Wheat Detection <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thanks, you're welcome :D</p> <p>Thanks! I used batch 1 and Google Colab (P100).</p> <p><a href="/nvnnghia">@nvnnghia</a> thanks :D</p> <p>Congratulations to all the winners and to all the players that learnt new things!</p> <p>Since the start of the competition we have used FasterRCNN then Shifted to EfficientDet then Shifted to YOLOv5. Then used a combination of both EfficientDet+YOLOv5. When we knew that YOLOv5 is prohibited we turned back to EfficientDet and kept optimizing it.</p> <p><strong>Training</strong><br> Data Cleaning: </p> <ul> <li>Deleted tiny bounding boxes (width or height under a certain threshold)</li> <li>Deleted hunge bounding boxes (width or height over a certain threshold)</li> </ul> <p>Data Augmentation: </p> <ul> <li>Online Augmentations</li> <li>Flips & Rotations (90 multiples)</li> <li>Crop & Resize</li> <li>Hue Saturation Contrast Brightness </li> <li>Moving Blur </li> <li>Noise Injection (Gaussian Noise)</li> <li>Cutout (Made different versions of that) </li> <li>Cutmix </li> </ul> <p>Training Process: </p> <ul> <li>Used EfficientDet5 (& EfficientDet7)</li> <li>Used both 512x512 and 1024x1024 images</li> <li>Startified 5 Folds Splitting </li> <li>By Source Splitting (for some Fine Tuning)</li> <li>Manual tweaks when training the models (& Early Stopping)</li> <li>Change Augmentation distributions every certain number of epochs</li> </ul> <p><strong>Inference</strong><br> A picture is worth a thousand words:</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1739631%2F7dffedb28611b84b7c76165e8873215d%2Fstar.png?generation=1596624192299446&alt=media" alt=""></p> <p>Here you find my inference <a href="https://www.kaggle.com/ffares/ensemble-models-tta-efficientdet-512-1024" target="_blank">kernel</a>, please upvote the notebook if you like it, and most imprtantly feel free to give your feedback so I can learn from my mistakes. </p> <p>I want to thank the hosts of this competition, kaggle and all the participants who submitted compliant solutions. Special thanks goes for <a href="https://www.kaggle.com/shonenkov" target="_blank">@shonenkov</a> <a href="https://www.kaggle.com/nvnnghia" target="_blank">@nvnnghia</a> <a href="https://www.kaggle.com/pestipeti" target="_blank">@pestipeti</a> and many others who kept sharing during the competition.</p> <p>We will update with further details.</p>
Global Wheat Detection	[Update ] The 9th place solution with code	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Global Wheat Detection <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Unfortunately, during the competition, we did not check the individual impact of this feature on the public score. I can only say that the metric on validation did not decrease. I'll write if I check it in the future.</p> <p>We didn't use Retinanet and YOLO to build the ensemble. We used DetectoRS and UniverseNet</p> <p>We trained models on 4/5 of the dataset and validated on 1/5 </p> <p>Here are some observations from our submissions: - Pseudo crops from jigsaw puzzles (DetectoRS R50): 0.7513 -> 0.7582 - Tuning of pseudo labeling parameters for sampling positive examples (ensemble): 0.7709 -> 0.7729 - Pseudo labeling (DetectoRS R50): 0.7582 -> 0.7691 - Pseudo labeling (UniverseNet Res2Net50): 0.7494 -> 0.7627 - SPIKE dataset (DetectoRS R50): 0.7582 -> 0.7592 - Deleting [usask1, ethz1] from pseudo labeling (DetectoRS R50): 0.7678 -> 0.7691</p> <p>For TTA was used a standard MMDet algorithm with NMS that looks like this for two-stage detectors (DetectoRS):<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2F6c66a60eb785874c2a8ab8a12bc9478c%2FSelection_126.png?generation=1596697435050357&alt=media" alt=""></p> <p>For one-stage detectors (UniverseNet), the algorithm is similar, only without the part with RoiAlign, Head, etc.</p> <p>WBF was used only for ensemble of two models.</p> <p>Thanks! </p> <p>For DetectoRS (one of the old models): 0.7582 -> 0.7691 For UniverseNet with Res2Net50: 0.7494 -> 0.7627</p> <p>Yes. I think therefore 1 epoch is enough.</p> <h1>Introduction</h1> <p>(This post was jointly written by the winning team, see details below)</p> <p>First, we wish to express our congratulations to the winners and to all the nominees as well for their remarkable achievements and continued excellence and look forward to exciting times in object detection algorithms.</p> <p>In addition, many thanks to Kaggle and the organizers for holding this interesting competition. And special thanks to my colleagues <a href="https://www.kaggle.com/orkatz2" target="_blank">@orkatz2</a> (Or Katz) and <a href="https://www.kaggle.com/solomonk" target="_blank">@solomonk</a> (Shlomo Kashani, author of the book “Deep Learning Interviews”).</p> <h1>Code</h1> <ul> <li>training code: <a href="https://github.com/amirassov/kaggle-global-wheat-detection" target="_blank">https://github.com/amirassov/kaggle-global-wheat-detection</a></li> <li>submission code: <a href="https://www.kaggle.com/amiras/pseudo-ensemble-detectors-3-st-universenet-r10" target="_blank">pseudo ensemble: detectors (3 st)+universenet r10</a></li> </ul> <h1>Summary</h1> <p>Our solution is based on the excellent MMDetection framework (<a href="https://github.com/open-mmlab/mmdetection)" target="_blank">https://github.com/open-mmlab/mmdetection)</a>. We evaluated and tested numerous models in an attempt to find the ones which are better suited for wheat detection. Why? Because there is not always a direct correlation between the COCO mAP score, as reported by the authors of the original models, and the actual mAP on the wheat corpus. Therefore, experimentation was extremely important. Amongst the many models we evaluated in MMDet are: </p> <ul> <li>GFocal</li> <li>ATSS</li> <li>UniverseNet </li> <li>DetectorRS </li> <li>SOLO-v2 (used for segmentation but we experimented with it anyway)</li> </ul> <p>We then trained an ensemble of the top-performing models which were: </p> <ul> <li>DetectoRS with the ResNet50 backbone (<a href="https://github.com/joe-siyuan-qiao/DetectoRS" target="_blank">https://github.com/joe-siyuan-qiao/DetectoRS</a>)</li> <li>UniverseNet+GFL with the Res2Net101 backbone (<a href="https://github.com/shinya7y/UniverseNet" target="_blank">https://github.com/shinya7y/UniverseNet</a>)</li> </ul> <p>To increase the score a single round of pseudo labelling was applied to each model. Additionally, for a much better generalization of our models, we used heavy augmentations.</p> <h1>Jigsaw puzzles</h1> <p>In the original corpus provided by the organizers, the training images were cropped from an original set of larger images. Therefore, we collected and assembled the original puzzles resulting in a corpus of 1330 puzzle images. The puzzle collection algorithm we adopted was based on <a href="https://github.com/lRomul/argus-tgs-salt/blob/master/mosaic/create_mosaic.py" target="_blank">https://github.com/lRomul/argus-tgs-salt/blob/master/mosaic/create_mosaic.py</a>. But we were unsuccessful in collecting the bounding boxes for puzzles. Mainly because of the existence of bounding boxes that are located on or in the vicinity the border of the image. For this reason, we generated crops for the puzzles offline in addition to training images and generated boxes for them using pseudo labelling.</p> <h1>Validation approach</h1> <p>We used MultilabelStratifiedKFold with 5 folds (<a href="https://github.com/trent-b/iterative-stratification" target="_blank">https://github.com/trent-b/iterative-stratification</a>) stratified by the number of boxes, a median of box areas and source of images. We guaranteed that there isn’t any leak between the sub-folds, so that the images of one puzzle were used only in that one particular fold.</p> <p>Referring to the paper <a href="https://arxiv.org/abs/2005.02162" target="_blank">https://arxiv.org/abs/2005.02162</a>, one can see wheat heads from different sources. We assumed that the wheat heads of usask_1, ethz_1 sources are very different from the test sources (UTokyo_1, UTokyo_2, UQ_1, NAU_1). Therefore, we did not use these sources for validation.</p> <p>However, our validation scores did not correlate well with the Kaggle LB. We only noticed global improvements (for example, DetectoRS is better than UniverseNet). Local improvements such as augmentation parameters, WBF parameters etc. did not correlate. We, therefore, shifted our attention to the LB scores mainly. </p> <p>We trained our models only on the first fold.</p> <h1>Augmentations</h1> <p>Due to the relatively small size of our training set, and another test set distribution, our approach relied heavily on data augmentation. During training, we utilized an extensive data augmentation protocol:</p> <ul> <li>Various augmentations from albumentations library (<a href="https://albumentations.ai):" target="_blank">https://albumentations.ai):</a><ul> <li>HorizontalFlip, ShiftScaleRotate, RandomRotate90</li> <li>RandomBrightnessContrast, HueSaturationValue, RGBShift</li> <li>RandomGamma</li> <li>CLAHE</li> <li>Blur, MotionBlur</li> <li>GaussNoise</li> <li>ImageCompression</li> <li>CoarseDropout</li></ul></li> <li>RandomBBoxesSafeCrop. Randomly select N boxes in the image and find their union. Then we cropped the image keeping this unified.</li> <li>Image colorization (<a href="https://www.kaggle.com/orkatz2/pytorch-pix-2-pix-for-image-colorization" target="_blank">https://www.kaggle.com/orkatz2/pytorch-pix-2-pix-for-image-colorization</a>) </li> <li>Style transfer (<a href="https://github.com/bethgelab/stylize-datasets)" target="_blank">https://github.com/bethgelab/stylize-datasets)</a>. A random image from a small test (10 images) was used as a style. </li> <li>Mosaic augmentation. <code>a, b, c, d</code> -- randomly selected images. Then we just do the following:</li> </ul> <pre><code>top = np.concatenate([a, b], axis=1) bottom = np.concatenate([c, d], axis=1) result = np.concatenate([top, bottom], axis=0) </code></pre> <ul> <li>Mixup augmentation. <code>a, b</code> -- randomly selected images. Then: <code>result = (a + b) / 2</code></li> <li>Multi-scale Training. In each iteration, the scale of image is randomly sampled from <code>[(768 + 32 * i, 768 + 32 * i) for i in range(25)]</code>.</li> <li>All augmentations except colorization and style transfer were applied online.<br> Examples of augmented images:</li> </ul> <table> <thead> <tr> <th><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2F54b1605a052bc34520cce5f7ca81f86f%2F0.jpg?generation=1596619282971124&alt=media" alt=""></th> <th><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2Fe7bad4c91782a5614ee5731f3610e376%2F6.jpg?generation=1596619343754665&alt=media" alt=""></th> </tr> </thead> <tbody> <tr> <td><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2F4dfab25fb6ed84e7ef545c42361382c5%2F27.jpg?generation=1596619635969807&alt=media" alt=""></td> <td><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2Fcb53ccda22bf16bef104caf9e25459df%2F47.jpg?generation=1596619666130434&alt=media" alt=""></td> </tr> </tbody> </table> <h1>External data:</h1> <p>SPIKE dataset: <a href="https://www.kaggle.com/c/global-wheat-detection/discussion/164346" target="_blank">https://www.kaggle.com/c/global-wheat-detection/discussion/164346</a></p> <h1>Models</h1> <p>We used DetectoRS with ResNet50 and UniverseNet+GFL with Res2Net101 as main models. DetectoRS was a little bit more accurate and however much slower to train than UniverseNet:</p> <ul> <li>Single DetectoRS Public LB score without pseudo labeling: 0.7592</li> <li>Single UniverseNet Public LB score without pseudo labeling: 0.7567</li> </ul> <p>For DetectoRS we used:</p> <ul> <li>LabelSmoothCrossEntropyLoss with parameter 0.1</li> <li>Empirical Attention (<a href="https://github.com/open-mmlab/mmdetection/tree/master/configs/empirical_attention" target="_blank">https://github.com/open-mmlab/mmdetection/tree/master/configs/empirical_attention</a>)</li> </ul> <h1>Training pipeline</h1> <p>In general, we used a multi-stage training pipeline:<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2F6b28b3ab6458d9a0763d34c78abf15a0%2FSelection_125.png?generation=1596629231468089&alt=media" alt=""></p> <h1>Model inference</h1> <p>We used TTA6 (Test Time Augmentation) for all our models:</p> <ul> <li>Multi-scale Testing with scales [(1408, 1408), (1536, 1536)]</li> <li>Flips: [original, horizontal, vertical]</li> </ul> <p>For TTA was used a standard MMDet algorithm with NMS that looks like this for two-stage detectors (DetectoRS):<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F413252%2F6fff06749b880222e0c06bea777b8e84%2FSelection_126.png?generation=1597392868005836&alt=media" alt=""></p> <p>For one-stage detectors (UniverseNet), the algorithm is similar, only without the part with RoiAlign, Head, etc.</p> <h1>Pseudo labelling</h1> <ul> <li>Sampling positive examples. We predicted the test image and received its scores and the bounding boxes. Then we calculated <code>confidence = np.mean(scores > 0.75)</code>. If the confidence was greater than 0.6 we accepted this image and used for pseudo labelling.</li> <li>Sources [usask_1, ethz_1] and augmentations like mosaic, mixup, colorization, style transfer weren’t used for pseudo labelling.</li> <li>1 epoch, 1 round, 1 stage.</li> <li>Data: original data + 3 x pseudo test data</li> </ul> <h1>Ensemble</h1> <p>We used WBF (<a href="https://github.com/ZFTurbo/Weighted-Boxes-Fusion" target="_blank">https://github.com/ZFTurbo/Weighted-Boxes-Fusion</a>) for the ensemble. The distribution of DetectoRS and UniverseNet scores is different. So we applied scaling using <code>rankdata</code> (<a href="https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.rankdata.html):" target="_blank">https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.rankdata.html):</a></p> <pre><code>scaled_scores = 0.5 * (rankdata(scores) / len(scores)) + 0.5. </code></pre> <p>WBF parameters:</p> <ul> <li>weights=[0.65, 0.35] respectively for models [DetectoRS, UniverseNet]</li> <li>iou_thr=0.55</li> <li>score_thr=0.45</li> </ul> <h1>Our final submission pipeline (0.7262 on Private LB and 0.7734 on Public LB):</h1> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F413252%2Fc2f34231c3dbffc0e8335b8d9cb15898%2FSelection_120.png?generation=1596619828315578&alt=media" alt=""></p> <h1>Some observations from our submissions:</h1> <ul> <li>Pseudo crops from jigsaw puzzles (DetectoRS R50): 0.7513 -> 0.7582</li> <li>Tuning of pseudo labeling parameters for sampling positive examples (ensemble): 0.7709 -> 0.7729</li> <li>Pseudo labeling (DetectoRS R50): 0.7582 -> 0.7691</li> <li>Pseudo labeling (UniverseNet Res2Net50): 0.7494 -> 0.7627</li> <li>SPIKE dataset (DetectoRS R50): 0.7582 -> 0.7592</li> <li>Deleting [usask1, ethz1] from pseudo labeling (DetectoRS R50): 0.7678 -> 0.7691</li> </ul> <h1>What we didn’t do:</h1> <ul> <li>MMDetection and YOLOV5 ensemble. That was the main goal but since YOLOV5 was disqualified we gave up this option.</li> <li>MMDetection and EfficientDet ensemble. We only managed to make a submission for the first time on the last day but we could not get a good enough result. We believe that if we had a few more days we could improve the result by tune the WBF parameters.</li> <li>We stopped improving YOLOV5 about a month before the end of the competition. </li> </ul> <h1>What didn’t work:</h1> <ul> <li>Wheat Ears Detection Dataset: CV improved, but LB did not.</li> <li>More than 1 rounds of pseudo labeling with the same model.</li> <li>Scale-aware testing like this: <a href="https://drive.google.com/file/d/14VwSjMeRZUtisZtqQPmbll6w4zvZXIAQ" target="_blank">https://drive.google.com/file/d/14VwSjMeRZUtisZtqQPmbll6w4zvZXIAQ</a></li> <li>Rotate90 TTA for DetectoRS and UniverseNet</li> </ul> <h1>Acknowledgements</h1> <p>Thanks to internet service provider HOSTKEY (<a href="https://www.hostkey.com" target="_blank">https://www.hostkey.com</a>) that gave me a grant for access to GPU servers.</p> <p><strong>The code will be published soon.</strong></p>
GoDaddy - Microbusiness Density Forecasting	#1 solution - generalization with linear regression	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>:( Not very kaggly, I think.</p> <p>Did you execute this code? Getting the mbd chg over data from revealed test should be fairly straightforward. </p> <pre><code>dfc = pd.read_csv() rt = pd.read_csv().set_index() dfc[] = dfc[].apply( x:x[-:]) dfc[] = dfc[].apply( x:(x[:x.index()])) dfc = dfc.reset_index().set_index() dfc[] = dfc.apply( r:(r[] - rt.loc[, ])/rt.loc[, ], axis = ) display(dfc.groupby().mean()[]) </code></pre> <p>I admit to being a bit surprised that most folks hadn't already done this sort of EDA. The contest will likely be won by whoever called the macro trend correct, which isn't always in a straight line.</p> <p>Thanks Nic! Good luck on your results.</p> <p>I predict all cfips. Manual blacklisting didn't seem to improve my scores, but I also had features which were similar in purpose, such as whether there was a change over some threshold in the last month(s). Rounding on active of course helps as well.</p> <p>Similar to my long trend prediction, though my may numbers might be a bit aggressive. My seasonal is very aggressive, but I figured why not in case things change dramatically. Also similar to the results I pulled from someone elses prediction.</p> <p>Just to be clear, the public leaderboard here is nonsense as it frequently is in most Kaggle contests. I enjoy overfitting them to show why that is the case. That said, there may be some advantage to leveraging january last values as I tried to do, but that remains to be seen. </p> <p>There will very likely be fairly massive shakeup (shake down) here and I fully expect to be among the latter group.</p> <p>For folks looking for a description of my model, I posted it here - <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/394810#2182222" target="_blank">https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/394810#2182222</a> </p> <p>However, my experience was similar to most folks - simple linear models were best, perhaps informed a bit by economic/census features, and there wasn't much to write about. The CV studies I did showed that using last split CV was frequently superior (though more volatile for sure) to using more extensive validation.</p> <p>New: I've uploaded the formal model submit file that was required to receive the prize distribution: <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/425000" target="_blank">https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/425000</a></p> <p>Code here - <a href="https://www.kaggle.com/kaggleqrdl/first-place-code" target="_blank">https://www.kaggle.com/kaggleqrdl/first-place-code</a>, see below for an explanation.</p> <p><strong>Acknowledgements</strong></p> <p>Let me first acknowledge Kaggle, the Venture Forward folks, and GoDaddy. Not just for creating Kaggle itself, hosting the contest and providing the microbiz data - incredible feats of faith, truly - but also for being patient with my fixation on transparency. I am an egalitarian thru and thru, and firmly believe transparency is what gets us there.</p> <p><strong>Introduction</strong></p> <p>The contest was to predict microbusiness density 3,4,5 months lookahead for mar/apr/may 2023 per each of the 3135 US counties in the dataset. Microbusiness density here is roughly equal to the # of GoDaddy <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/374293" target="_blank">internet domains registered</a> / adult population in a county as measured by a 2 year lagging US census. </p> <p>Eg: if a total of 10 GoDaddy domains were registered in <a href="https://en.wikipedia.org/wiki/De_Kalb,_Missouri" target="_blank">De Kalb, MO</a> with 200 adult population according to the 2021 census in 2023 February, than microbusiness density for that county would increase by 10/200 or 0.05 in that month.</p> <p>We were given historical microbiz data for 2019 August -> 2022 December to train with, and a not-so-secret 'public LB' of data for January to validate against. Scores on the public LB were provided by Kaggle, and we were given 5 submissions per day.</p> <p><strong>Method</strong><br> To win this, I took a baseline prediction for the january public leaderboard based on overfitting it, and used linear regression to project into mar/apr/may</p> <pre><code>, [, , ,] - month lookahead , [, , ,] - month lookahead , [, , , , ] - month lookahead </code></pre> <ol> <li>l# here means lag # of months, shift is the actual value, just l#_ is pct_change</li> <li>shiftf1_l1_active_logit is whether there was a change in active over previous month (l1_active != 1)</li> <li>pct_college_2021 / pct_bb_2017 is from the census data the hosters shared</li> <li>uemp / lf / ur are from the dataset I shared here - <a href="https://www.kaggle.com/datasets/kaggleqrdl/gd2022datasets" target="_blank">https://www.kaggle.com/datasets/kaggleqrdl/gd2022datasets</a></li> </ol> <p>I selected these features by cross validating on the last possible window (2,3,4 month respectively) without overlap. </p> <p><strong>Insights</strong><br> The key ideas that seemed to help:</p> <ul> <li>LB overfitting / probing. I did this by using the top public LB notebook plus focusing on the 10 or so counties that had recent significant volatility in the dataset. This was done because in non stationary time series data, the last value is your most powerful and compelling signal. It's worth noting that due to the <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/413974" target="_blank">dec22/jan23</a> snafu (we were given 60% incorrect data, as measured by SMAPE), this technique was severely impaired and my lead in the public LB went from ~0.2 to ~0.08 over the top 20. Overfitting the LB doesn't work very well when the LB has wrong data. <br> More explanation here on this approach - <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/413109" target="_blank">https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/413109</a> I also explained these ideas early on in the contest <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/375244" target="_blank">here</a> and <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/378002" target="_blank">here</a>. The <a href="https://www.kaggle.com/kaggleqrdl/first-place-code" target="_blank">notebook</a> has code used for SMAPE root solving and probing techniques</li> <li>LinearRegression - I tried practically every() other model and techniques, even dl/nn. LR consistently resulted in more reliable CV scores. Overfitting is always a concern with time series data and it's so easy to tune other models and convince yourself you found something superior.</li> <li>Early stopping on greedy feature selection. This was done mostly on an intuition basis, but after running 1000s of experiments I developed a good feel for this I believe. I noticed overfitting occurring when I did more than 4 or 5 features, plus the decreased rate of improvement was another signal that overfitting had occurred. Certain features as well seemed to have more signal than others (eg, L10/11 employment, bb census) so I would keep going until they made it into the list but not much more.</li> <li>Last window CV for feature selection. My hypothesis is that as a non stationary time series the relevant predictive features change and it's best to use the latest CV window (last few months of 2022). This played out in the results I saw over the experiments I did, but I will say std deviation on the error term was quite high, even if the mean was superior to longer CV periods. The risk was that so little CV time made it a bit of a gamble.</li> <li>Use as much original training data as possible - Generalizing through data sufficiency was a reoccurring theme in this comp, and any technique that didn't appreciate it was degraded respectively. I used everything up to the lag window as per above. I didn't modify any of the data (beyond removing 3 problematic counties), even left the break in 2021 alone. I tried doing all sorts of things to clean up that data, but they just seemed to degrade results. </li> <li>rounding based on active/population of course was a must. I found clipping, in the end, didn't help.</li> </ul> <p>every regression model in sklearn, various arima libraries (a lot of time here, wasted I feel), hierarchical, xgb, catboost, lgbm, pytorch/dll. I tried both manually tuning the previous as well as various auto approaches like grid search/optuna. Lots of very exciting results, but eventually it occurred to me I was probably just overfitting.</p> <p>It's worth observing that my lead against the #2 and #3 spots increased significantly from the public LB versus the private LB. I believe this was due to superior predictions for apr/may as my march results were somewhat lackluster. My lead over the top 20 spots also increased, though it is less clear how much of the public LB influenced their models.</p> <p><strong>Ideas that didn't seem as important</strong></p> <p>I also tried a lot of different datasets (including all the ones I shared publicly) up to and including google keyword trends for 'godaddy'. :) The above worked best when combined with early stopping. There would be sparks of good CV scores when using other features, but adding them frequently felt like overfitting. </p> <p>I briefly looked into using zone files and whois databases, but these were not openly available so of course I didn't use them. </p> <p>Crawling/Scraping would have in theory been public data, but I had committed to myself not to use any data that I didn't share openly, so I figured it wouldn't be worth the effort required. </p> <p>Everything I used I shared in the <a href="https://www.kaggle.com/datasets/kaggleqrdl/gd2022datasets" target="_blank">dataset</a> I made publically available at the beginning of the comp.</p> <p>I did a lot of initial work around per county models, work I eventually decided was wasted. There was barely enough data to validate the models and features above. There might be a way of segregate the counties, the question is - Would you have to segregate your training data as well? Would the increased complexity be worth what might potentially be only incremental improvements? Hard to say, especially since observing incremental improvements is mostly impossible when your CV scores are so volatile. </p> <p><strong>Future directions</strong><br> One thing I spent time looking at was correlation between the counties. There is an argument to be made I think that different counties may follow similar trends at potential lag times. There might be more to do here, but it would need to be done globally rather than individually, and your models would require appropriate data sufficiency. </p> <p>Another feature I didn't plumb carefully enough was 1/2/3/etc annual renewal on GoDaddy domains, an idea I got by chatting with some domain resellers recently on a reseller domain community forum. I suspect they would provide a well of ideas around this that could be leveraged more.</p> <p><strong>Final thoughts</strong></p> <p>As I mentioned above, and others have mentioned, the data was non stationary except for some global linear growth coefficients. It's also been said that the smaller counties added a lot of noise, but I believe it was the right choice to try to see if we could do something there. There is a paucity of stats and analysis around more rural areas and emergent issues are hard to detect because of it.</p> <p>Another way of looking at the data might be volatility instead of direction. The wild fluctuations in active domains could just be due to data errors but they could also be due to underlying social and economic factors. In the latter case, we need to consider: could the volatility be an opportunity, like the first sparks of a fire? Could these counties benefit from more targeted and opportune investment? If volatile counties have increased potential, what are the triggers for volatility? These are questions and experiments (signal + intervention) worth considering</p> <p>If the VF folks would like to contact me (heh), feel free to reach out. As you may have noticed, I could talk about this stuff all day long with folks who are equally interested.</p> <p>..</p> <p>There was a bug in the team update that didn't let me change the LB pointer which I had originally pointed to the message below, so you're stuck with both. I originally pointed it here as I was more interested in what other folks did as I wasn't really that optimistic about my simplistic approach above. In retrospect, I probably should have trusted the process of elimination that I went through.</p> <p>previous thread title: <strong>What is your mean active change over december for mar/apr/may?</strong></p> <p>Edit - to make clear up front, you can share your results from your submission by just executing this code. It uses microbusiness density and not active, but that should be fine for comparison.</p> <pre><code>dfc = pd.read_csv() rt = pd.read_csv().set_index() dfc[] = dfc[].apply( x:x[-:]) dfc[] = dfc[].apply( x:(x[:x.index()])) dfc = dfc.reset_index().set_index() dfc[] = dfc.apply( r:(r[] - rt.loc[, ])/rt.loc[, ], axis = ) display(dfc.groupby().mean()[]) </code></pre> <p>….</p> <p>It's going to be awhile before we get final results, and I think May results might play a large role in determining the eventual winners. While we wait, it might fun to share what we actually predicted. We could compare our actual results, which would be good and I've uploaded my two <a href="https://www.kaggle.com/datasets/kaggleqrdl/gd2022-subs" target="_blank">here</a> - but for those who don't want to and perhaps simpler might just be sharing mean % change over active.</p> <p>Here's the code I used. It requires having accurate values for december in your dataframe and also have calculated 'active' as one of your columns. </p> <p>If you don't want to use active, you can also share microbusiness density as well, but I found it less interesting because of the pop shift. There were some outliers in the pop shift that made it harder to ensure that nothing went awry, so I liked having active available. I've shared both below.</p> <p>You can calculate cfips/first day with this code:</p> <pre><code>dfc[] = dfc[].apply( x:x[-:]) dfc[] = dfc[].apply( x:(x[:x.index()])) </code></pre> <p>Calculate active_chg values with this code:</p> <pre><code>dfc = dfc.reset_index().set_index() dfc[] = dfc.apply( r:(r[] - dfc.loc[, ])/dfc.loc[, ], axis = ) dfc.groupby().mean()[] </code></pre> <p>This is my seasonal model </p> <pre><code>first_day_of_month --01 - --01 -01-01 -02-01 -03-01 -04-01 -05-01 Name: active_chg, dtype: float64 </code></pre> <p>This is my long trend</p> <pre><code>first_day_of_month --01 - --01 -01-01 -02-01 -03-01 -04-01 -05-01 Name: active_chg, dtype: float64 </code></pre> <p>Using mbd code if you didn't have an active column:</p> <pre><code>dfc = dfc.reset_index().set_index() dfc[] = dfc.apply( r:(r[] - dfc.loc[, ])/dfc.loc[, ], axis = ) dfc.groupby().mean()[] </code></pre> <p>Seasonal sub</p> <pre><code>first_day_of_month --01 - --01 -01-01 -02-01 -03-01 -04-01 -05-01 Name: mbd_chg, dtype: float64 </code></pre> <p>Long trend sub</p> <pre><code>first_day_of_month --01 - --01 -01-01 -02-01 -03-01 -04-01 -05-01 Name: mbd_chg, dtype: float64 </code></pre> <p>Note for february I just used the january values. Sharing january isn't really that helpful of course, I just do it for fun here.</p> <p>I'll answer if you share your averages ;p <a href="https://www.kaggle.com/judith007" target="_blank">@judith007</a> </p> <p>I pulled this from another contestants submission that they shared publicly. I won't give any names, but I found it interesting</p> <pre><code>-03-01 -04-01 -05-01 </code></pre> <p>Thanks Chen. I was hoping some folks might share their results, but so far everyone seems a bit shy. Ah well.</p> <p>Thanks Luca, good luck to you as well.</p> <p>Actually, I did provide some code above for just using microbusiness density. </p> <p>Perhaps I should make this more clear.</p> <p>If you run this, it should work. </p> <pre><code>dfc = pd.read_csv() dfc[] = dfc[].apply( x:x[-:]) dfc[] = dfc[].apply( x:(x[:x.index()])) dfc = dfc.reset_index().set_index() dfc[] = dfc.apply( r:(r[] - dfc.loc[, ])/dfc.loc[, ], axis = ) display(dfc.groupby().mean()[]) </code></pre> <p>Yes, it does require having accurate december values. The pop shift problem should be OK as we'll all have it for comparison.</p> <p>You can use this, I'll add it above</p> <pre><code>rt = pd.read_csv().set_index() dfc[] = dfc.apply( r:(r[] - rt.loc[, ])/rt.loc[, ], axis = ) </code></pre> <p>Well, I let it run a little wild because of the layoffs in the tech sector. I sort of felt those folks might try to start up microbiz. Historically the tech sector has seen full employment, but lately less so, so I think using historical values we might be seeing some underestimations.</p> <p>It's all guesses and fun though. I suspect we'll see a lot of new folk getting gold here</p> <p><a href="https://www.kaggle.com/sireesh" target="_blank">@sireesh</a></p> <p>Can you share them? Did you use fairly simple models?</p>
GoDaddy - Microbusiness Density Forecasting	11th Place Solution - Simplicity (Luck) Is All You Need	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Ty, I hope it would be a gold for me😂.</p> <p>Do you mean using the data from <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/411428" target="_blank">here</a>?</p> <p>I got </p> <pre><code>. for single . for </code></pre> <p>First, my thanks to Kagglers for very informative and helpful discussions and notebooks. Also, thank you to Kaggle and GoDaddy for hosting the competition.</p> <p>My code is available at <a href="https://github.com/ywugwu/GoDaddy-MDF-Competition" target="_blank">github</a>, which is based on <a href="https://www.kaggle.com/code/vadimkamaev/microbusiness-density" target="_blank">VADIM KAMAEV's</a> public notebook. </p> <p>In this competition, my idea is to keep improving my CV and PB score without letting my model grow too complex.</p> <h1>Validation (CV)</h1> <p>I used the last 5 month data for validation. The model with the best validation score is used as one of my final submissions.</p> <h1>Model</h1> <p>Same with the public notebook, I used a Catboost Regressor to stack a LightGBM, an XGBoost, and a CatBoost model. Tuning a lot hyperparameters.</p> <h1>Feature Engineering</h1> <p>I didn't do too much novel feature engineering, only used techniques from public notebooks. </p> <p>But I change the hyperparameters a lot.</p> <h1>Blending Submissions</h1> <p>I used different hyperparameters of the model and different features (i.e., with different diff, max, mean, etc. settings) to conduct a large number of submissions. I select the best 4 (or maybe 3) submissions and use a weighted average to create a blended submission. The best (according to the public leaderboard) blended result is selected as one of my final submission. In the end, the private leaderboard score of blended submission is slightly better than the best CV model.</p> <p>thanks for your reminder, I added an extra <code>.</code> at the end of my URL 😭. Now, It should work. </p>
GoDaddy - Microbusiness Density Forecasting	2nd Place solution: The Godaddy Microbusiness Data Cleaning Challenge	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>You probably would have gotten a similar result based on your technique. </p> <p>For the cross validation, I know in some public notebooks there was an explicit violation of proper CV technique since the smoothing was done before the cross validation. I think to properly setup your environment given the challenge and the data quality, all data cleaning, etc should be a part of the model itself and all cross validation should be done in a walk forward mode since it is a time series challenge. </p> <p>For example, the GRU models used with Group K fold would in this case use the the group CV for early stopping and validation of the training phase, but to test the overall model one would need to use the walk forward method.</p> <p>Not sure of your setup, so this may not affect you</p> <p>For the Jan predictions, not sure I can answer that until the end of the competition. Much of that is just a hope that the interaction of those predictions with my model will not cause some bias that I could not detect.</p> <p>Given my CV setup, a last value model gave these results on the original training data:<br> Train up to and including data from 2022-01-01 and evaluate forecasts from 2022-04-01 to 2022-06-01 = 3.1956774734784967<br> Train up to and including data from 2022-02-01 and evaluate forecasts from 2022-05-01 to 2022-07-01 = 3.4152913611520037<br> Train up to and including data from 2022-03-01 and evaluate forecasts from 2022-06-01 to 2022-08-01 = 3.3317878979931765<br> Train up to and including data from 2022-04-01 and evaluate forecasts from 2022-07-01 to 2022-09-01 = 3.3842283789667764<br> Train up to and including data from 2022-05-01 and evaluate forecasts from 2022-08-01 to 2022-10-01 = 3.7841915766675602<br> Average SMAPE = 3.4222353376516024</p> <p>And the final model I used:<br> Train up to and including data from 2022-01-01 and evaluate forecasts from 2022-04-01 to 2022-06-01 = 2.962060956397758<br> Train up to and including data from 2022-02-01 and evaluate forecasts from 2022-05-01 to 2022-07-01 = 3.097867393456829<br> Train up to and including data from 2022-03-01 and evaluate forecasts from 2022-06-01 to 2022-08-01 = 3.0168450673149243<br> Train up to and including data from 2022-04-01 and evaluate forecasts from 2022-07-01 to 2022-09-01 = 2.9881429431194393<br> Train up to and including data from 2022-05-01 and evaluate forecasts from 2022-08-01 to 2022-10-01 = 3.3630555199321344<br> Average SMAPE = 3.085594376044217</p> <p>Interesting problem, and it was also interesting to see the solutions came up with. I'm pretty new to these competitions. The biggest uncertainty to me is how much probing of the hidden data is allowed, especially in a case like this where it matters a lot for time series prediction.</p> <p>I also though SMAPE was a weird choice of metric that didn't make sense in a business sense. A change in a small county would have a large influence in SMAPE, where if you are trying to capture revenue streams, wouldn't you care a lot more about a change in the number of active forecasts in Los Angeles? I can only imagine this would be useful for allocation of advertising dollars.</p> <p>The code: <a href="https://www.kaggle.com/code/danielphalen/godaddy-microbusiness-final/notebook?scriptVersionId=133942554" target="_blank">Godaddy Microbusiness Final</a></p> <h2>The biggest issue - Data quality</h2> <p>As has been pointed out by many people, SMAPE is a relative metric. With the kind of forecast values we are looking at (~1.2-1.5 for 1 month, ~3.2 for months 3-5 together), we can look at the contribution of an individual CFIPS. </p> <p>$$\Delta_{SMAPE,i} = \frac{200}{n_{counties}} \frac{\|F_i - A_i\|}{F_i+A_i}$$</p> <p>Which, in the case of something going from zero to non-zero can be ~0.0638 per month difference. At the same time, a prediction for most of these models seems to be on the order of 0.5% change per month, which is far below the smallest change in active entries for many 25% of the counties. </p> <p>Second, there are many CFIPS where the data is terrible. The hosts acknowledged that they had a methodology change in Jan 2021, leading to a number of jumps, and I suspect there was another change after the first month. </p> <p>The one which always bothered me was CFIPS 56033, Sheridan Co, WY, where there are 2.36 microbusinesses/working age person. I can only guess there is some bug where if there is a misclassification they dump it in that CFIPS. I also wondered if there was some fraud happening during COVID times as people chased PPP loans given the large jumps. However, many of these issues would eventually revert, like the below:</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5552115%2Fc85777fd4f981dee2765e2b78b7f1236%2Fcfips48155.png?generation=1678978908088571&alt=media" alt=""></p> <p>So the question I think that make or breaks this, if we identify a large jump, will it revert? In this I believe we are helped by the gap between December and the first forecast month, March. I also hope GoDaddy gets better in their data collection procedures, which would help. In then end, if they want to use this in business, it would probably not be that useful to have a large machine learning error correction model for data collection issues.</p> <p>My solution is a mix of public leaderboard probing for individual CFIPS changes, reversion for outlier CFIPS, and a forecast of the smooth changes. </p> <h3>Public leaderboard - still very useful information</h3> <p>Given the equation above and as <a href="https://www.kaggle.com/petersorenson360" target="_blank">@petersorenson360</a> suggested, and that the leaderboard was active to 4 decimal places, you are sensitive to changes of less than 1% in the value an individual value. Given you had about 20 days from revealed test to final submission deadline, you could probe about 100 CFIPS for final values. It was actually easier if you made the value worse since the equation above would allow you to work out the exact number of active entries. </p> <p>Many large jumps would the revert, which we tried to work out how long that might take. This will probably be the difference in the end of who wins and loses, and I think the couple month gap between the test and private data will help ensure they revert. This was by far the top influence.</p> <h3>Continuous Model</h3> <p>After seeing GiBa's <a href="https://www.kaggle.com/titericz" target="_blank">@titericz</a> notebook, his data cleaning method reminded me of something used in futures algorithmic trading, called a <code>continuous contract</code>. Basically, you need to take a discontinuous series of prices and make it smooth. So for what I will call the continuous forecast, I used a data cleaning method where I looked for large jumps in the number of active entries, then did a shift to smooth them out. So the month where there was a large active jump became zero active jump. This smoothing method gave the best CV score of the number that I tried.</p> <p>I then setup the CV environment. For the model used XGBoost, added extra indicators, and prevented peeking in the future:</p> <ul> <li>lagged density changes and lagged active values</li> <li>pct_bb, pct_college, pct_it, pct_foreign_born, median_hh_inc of the last year (Many public books used an implicit forward bias in their features by looking at 2021 census data when training on 2019 data)</li> <li>Labor Force and Labor force participation for the county</li> <li>10 year average pct population change for the county</li> <li>latitude and longitude</li> <li>engagement, participation, and MAI_composite from the Godaddy website.</li> <li>The difference between the microbusiness density and the average of the neighboring counties, weighted by population.</li> </ul> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5552115%2Ff1ba9755aeb2b45525fd98359dbf9cd2%2FFeatureImportances_1.png?generation=1687036241940575&alt=media" alt=""></p> <p>I found a population cut of about 5,000 was helpful. Rounding the model to an integer number of active helped a bit. A lot of these external indicators proved more helpful for the longer term forecasts, where there is some reversion to an average.</p> <p>Other notes:</p> <ul> <li>I actually found training on the 1 month forward change then feeding that prediction back into the model and recalculating all indicators gave the best CV, as opposed to directly forecasting the 3 month ahead forecast. It turned out to be less biased.</li> <li>The <code>scale</code> variable used seemed to work in forecasting the public leaderboard, but did terribly in my CV environment.</li> <li>Again, noting that the last value is pretty good and I was using a roll forward model, I used the tuned public models to patch in a January forecast as an estimate for the ground truth, then rolled forward the XGBoost model from that.</li> </ul>
GoDaddy - Microbusiness Density Forecasting	35th Place Solution - GRU+XGB	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Goal of this competition was to predict monthly <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/374293" target="_blank">domain registrations of small businesses</a> divided by the active population for all counties in the US. The biggest difference to other timeseries competitions is in my opinion that the public leaderboard tracked the score of the one month forecast (January) while our final score would be determined on 3-5 months in the future (March, April, May). My expectation for this competition was to deal with various macroeconomic explanatory variables and maybe even include some capital markets data. As it turned out this was not the case😄.</p> <h3>Things that did not work</h3> <ul> <li>I spent more time than I'm willing to admit on scrolling through files from statsamerica (<a href="http://www.statsamerica.org/downloads" target="_blank">www.statsamerica.org/downloads</a>) trying to find anything helpful to enhance the data we had at hand. Unfortunately I can't present any evidence for it's usefulness in this context. Anyway they have cool stuff (<a href="https://www.statsamerica.org/innovation" target="_blank">https://www.statsamerica.org/innovation</a>) over there.</li> <li>Capital markets indicators also had little effect. The decision to register a domain is surprisingly not driven by the VIX or FOMC dots (or more precise it does not help on county level forecasting).</li> </ul> <h3>Data Quality and Preprocessing</h3> <p>Instead, outlier detection was the name of the game. I didn't probe the leaderboard but nevertheless spent much time comparing small counties and their erratic timeseries. Additionally, we had to deal with a <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/375715" target="_blank">structural change</a> in the data in January 2021 and the <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/389215" target="_blank">underlying population numbers</a> were also shifting. </p> <p>Therefore the target variable was adjusted for the latest census data and I removed suspect jumps for all counties in this month if they exceeded a threshold of +-7%:</p> <pre><code> o tqdm(raw.cfips.unique()): indices = (raw[]==o) tmp = raw.loc[indices].copy().reset_index(drop=) var = tmp.microbusiness_density.values.copy() var_pct = tmp.microbusiness_density.pct_change().clip(-,) j (, , -): j== (var_pct[j]== - var_pct[j]== ): var[j-] = var[j] : var[j-] = var[j]/(+var_pct[j]) raw.loc[indices, ] = var </code></pre> <p>Afterwards the timeseries was further denoised with an autoencoder:</p> <pre><code> (): i = tf.keras.Input(shape=(,)) encoded = tf.keras.layers.BatchNormalization()(i) encoded = tf.keras.layers.GaussianNoise(noise)(encoded) encoded = tf.keras.layers.Dense(,activation=)(encoded) decoded = tf.keras.layers.Dropout()(encoded) decoded = tf.keras.layers.Dense()(decoded) x = tf.keras.layers.Dense(,activation=)(decoded) x = tf.keras.layers.BatchNormalization()(x) x = tf.keras.layers.Dropout()(x) x = tf.keras.layers.Dense(,activation=)(x) encoder = tf.keras.Model(inputs=i,outputs=x) loss = tf.keras.losses.MeanSquaredError() encoder.(optimizer=Adam(),loss=loss) encoder </code></pre> <p>I'm unsure if this could add any kind of leakage as it was applied before/outside of the CV fold structure. But it stabilized my CVs and had no negative impact on the public leaderboard score. Anyway I kept using it only for the GRU model and trained XGB on the 'raw' data.</p> <h3>Models and Target</h3> <p>Only counties with more than 150 microbusinesses were considered as input for the model. For counties below this threshold the last value was used.<br> Building on top <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/381038#2114357" target="_blank">Chris proposal</a> I used GRU with 24 months sequence of np.log1p(microbusiness density) input combining 3 folds (GroupKFold grouped by date) and forecasting 1, 3, 4, 5 months into the future.<br> As a second model I used XGB derived from what <a href="https://www.kaggle.com/code/titericz/better-xgb-baseline" target="_blank">GIBA shared</a>, defining the 1, 3, 4, 5 months growth difference as a target. Most of the county level features were included here but I think they added little explanatory value. For postprocessing I clipped the XGB predictions of the percentage changes with the corresponding 10% and 90% quantile values observed after preprocessing. Both models are weighted with 50% in the final submission.</p> <h3>Second Submission</h3> <p>This was a simple linear model inspired by the success of <a href="https://www.kaggle.com/code/vitalykudelya/21-lines-of-code" target="_blank">21 lines of code</a>. It calculated the average monthly growth per county on the 21 months after the structural break in January 2021. The mean was clipped at -0.5% and +1% and for counties with less then 150 active domains this value was scaled by 0.3 (all of this was expert judgement based on very crude statistics, average growth, 5%,25%,75%,95% quantiles of growth). For counties with stale growth in November and December the last value was forecasted perpetually. This scored 4.0731 which would have been place 175 (still a silver medal with 10% of the effort). 🙀</p>
GoDaddy - Microbusiness Density Forecasting	3rd Place - Predict Multipliers with GRU	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><blockquote> <p>How is the 1.0045 value obtained through probing used? When predicting 5 values, is it fixed as the first ratio value?</p> </blockquote> <p>After the GRU makes prediction for January, we compute the ratio of "average January prediction" divided by "average December prediction" for large counties. The result is <code>1.0032</code>. This is not the optimal ratio. So we multiply every January prediction for large counties by <code>1.0012</code>. Afterward, the the ratio of "average January prediction" divided by "average December prediction" is <code>1.0045</code> and our public LB and private LB are boosted.</p> <p>Regarding the small counties, each begins with multiplier of <code>1</code> which means that each small county prediction uses the December 2022 value. We then try different multipliers for different small counties and keep ones that improve LB. Basically we are finding what the small counties ground truth is for January 2023. Then our final submission uses the January values for small counties instead of December values for small counties. This boosts public and private LB.</p> <p>Both post processes combined improve CV and LB by about <code>+0.05</code></p> <p>Thank you Kaggle and GoDaddy for an exciting forecasting competition!</p> <h1>Predict Multipliers with TensorFlow GRU</h1> <p>Our forecasting models need to predict 3135 county microbusiness density predictions per future month. The public LB is one future month and the private LB is three future months (with 2 month gap between train and test).</p> <p>From Kaggle's M5 competition <a href="https://www.kaggle.com/competitions/m5-forecasting-accuracy/discussion/163621" target="_blank">here</a> we learned that predicting multipliers (between the test time period and train time period) was most important. For example in M5, we could take an average public notebook and multiply all predictions by <code>0.95</code> and win Gold medal !! Similarly in Kaggle's GoDaddy competition, a simple last value baseline multiplied by the correct January/December multiplier of <code>1.0045</code> as shown in Vitaly's notebook <a href="https://www.kaggle.com/code/vitalykudelya/21-lines-of-code" target="_blank">here</a> achieved Gold medal public LB for many months!</p> <p>To predict multipliers, we convert train data into multipliers and then train a TensorFlow GRU. This simple model achieves 15th place Gold. Then if we post process the multipliers based on ratios learned from LB probing, we boost our solution to 3rd place Gold !!</p> <h1>Original Train Data</h1> <p>Kaggle gave us 41 historical months of data for each of 3135 USA counties. </p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Jun-2023/data1.png" alt=""></p> <h1>Step 1 - Adjust Train Data</h1> <p>The definition of microbusiness density is <code>micro businesses per 100 people over age 18</code>. The value of microbusiness density changes when county population changes. So the first step is to convert all microbusiness density to use the same 2021 census. We use the formula</p> <pre><code> = microbusiness_density * (population / population_year_2021) </code></pre> <h1>Step 2 - Create 56,000 Time Series!</h1> <p>For each county, we create 18 time series. We use 13 consecutive months to train our GRU and predict the next 5 months. (Pictured below is only 3 time series per county for illustration purposes). Afterward, we have 56,000 time series to train with!</p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Jun-2023/data2.png" alt=""></p> <h1>Step 3 - Use Only Top 90% Largest Counties</h1> <p>We notice that the smallest 10% counties have nearly the same microbusiness density from month to month, so we only train our GRU using the largest 90% counties. During inference, we will use the last known value as prediction for small counties.</p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Mar-2023/growth.png" alt=""></p> <h1>Step 4 - Convert to Multipliers</h1> <p>Each county has 13 months of train and 5 months of valid. We will convert the raw micro business data into ratios by replacing each value with the ratio of current month divided by previous month. Afterward we will have 12 months of train ratios and 5 months of valid ratios</p> <pre><code> k in range(): [:,k+] = old_data[:,k+] / old_data[:,k] </code></pre> <h1>Step 5 - Train GRU with GroupKFold</h1> <p>When creating KFold we need to use GroupKFold and keep all time series for each county (from the possible 18 time series per each county) within the same fold. Otherwise CV score will be inaccurate and we cannot optimize our models hyperparameters correctly. Below is our TensorFlow GRU model which takes an input of 12 ratios and predicts 5 ratios:</p> <pre><code>def build_model(): inp = tf.keras.Input(shape=(12,1)) # INPUT SHAPE IS 12 x = tf.keras.layers.GRU(=8, =)(inp) x = tf.keras.layers.GRU(=8, =)(x) x = tf.keras.layers.GRU(=8, =)(x) x = tf.keras.layers.Dense(5,=)(x) # OUTPUT SHAPE IS 5 model = tf.keras.Model(=inp, =x) opt = tf.keras.optimizers.Adam(=1e-4) loss = tf.keras.losses.MeanSquaredError() model.compile(=loss, optimizer = opt) return model </code></pre> <h1>Step 6 - Infer and Post Process</h1> <p>We predict each county individually. To make predictions for January 2023. We begin with the last known value (of the county we are predicting) from December 2022 and multiply by the first ratio predicted by our model. To predict Feb 2023, we take our Jan 2023 prediction and multiply by the second ratio predicted by our model. To predict Mar 2023, we multiply Feb 2023 by third ratio. To predict Apr 2023, we multiply by fourth ratio. And finally to predict May 2023, we multiply by fifth ratio.</p> <p>We can improve our predictions by modifying the first ratio (which is the ratio of January 2023 divided by December 2022) by information from probing the public LB. Probing informs us that the best January/December average ratio for the largest 90% counties is <code>1.0045</code>. Similarly, we can improve our predictions for small counties by probing the public LB to find ratios for different sized small counties.</p> <p>Post process boosts our GRU solution from 12th place Gold to 3rd place Gold !!</p> <h1>GRU Solution Code Published</h1> <p>Full solution code with preprocess, train, and infer is published <a href="https://www.kaggle.com/code/cdeotte/gru-model-3rd-place-gold" target="_blank">here</a>. </p> <h1>More Solution Models</h1> <p>My final two submissions were</p> <ul> <li>GRU with PP - 3rd place</li> <li>Linear Model with PP - 10th place</li> </ul> <p>More information about my linear model is <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/395098" target="_blank">here</a>. Enjoy!</p> <p>No, my previous post is <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/395098" target="_blank">here</a></p> <p>Congratulations <a href="https://www.kaggle.com/kaggleqrdl" target="_blank">@kaggleqrdl</a> winning first place!</p> <p>The competition metric is SMAPE which is about ratios of error divided by mean of prediction and ground truth. When training a model, using different losses can have different effects on model accuracy with relation to competition metric. Another way to affect model accuracy without changing the loss function is to transform the targets. Using MSE loss with ratios does a better job optimizing SMAPE than trying to predict the raw microbusiness values with another loss. Also transforming targets make more targets similar. For example when using raw microbusiness values, maybe the values do not repeat themselves often, but when converting everything into ratios, the ratios do repeat themselves often which reinforce the model's learning.</p>
GoDaddy - Microbusiness Density Forecasting	44th Place Solution - Simple LGBM	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thank you for the competition on a very interesting topic. It was a time-series forecasting problem that predicts the true future, and was a fun theme to solve. I understood that the United States is a very large and diverse country.</p> <p>This is a simple LightGBM, 44th Place Solution.</p> <h3>Solution Overview</h3> <ul> <li><p>Only LightGBM with 32 features　</p></li> <li><p>Multiplier Prediction<br> I chose target as the multiplier between the previous month's data and the month to be predicted.</p></li> <li><p>Target smoothing<br> Because there were so many outliers in the data, I smoothed the target by taking the medians of the three cases before and after the target.</p></li> <li><p>Model to forecast 1-6 months later<br> I created several models to forecast 1-6 months later separately.</p></li> <li><p>Average of 3 months<br> Average the multipliers for 3 months, including the months before and after the month of forecast. This is more accurate.</p></li> <li><p>Conversion of population from 2020 to 2021</p></li> <li><p>Round(0) since it is a discrete value</p></li> </ul> <h3>My Notebook</h3> <p>I have published my notebook below.<br> <a href="https://www.kaggle.com/code/thajime/godaady-44th-solution-lgbm" target="_blank">https://www.kaggle.com/code/thajime/godaady-44th-solution-lgbm</a></p>
GoDaddy - Microbusiness Density Forecasting	62th Place Solution - Last Value and Rule-based Multiplier	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>I selected constants that were minimizing my validation score so I didn't think I was overfitting at any time. </p> <h2>Validation</h2> <p>Validation is really important for problems like this. I baffled by lots of people validating on last month without any gap. Replicating test set split was essential for the validation scheme so I used 8 folds of 2 months gap and 3 months validation with expanding window over the entire dataset. I also added 3 additional folds with their training period ending on December for simulating new year transitions. I had 11 folds but I didn't necessarily use all of them at the same time.</p> <h2>Models</h2> <p>I tried some models but none of them improved my validation score so I had to use last value baseline. These are the models and techniques I tried:</p> <ul> <li>Linear regression, LightGBM and XGBoost with raw and external pivoted features from GoDaddy Venture Forward Data Hub</li> <li>Linear regression with normalized month and target for each county</li> <li>ARIMA, SARIMAX and other auto-regressive model variations for each county</li> <li>Directly optimizing a multiplier for last value using bayesian optimization</li> <li>Last value and a ruled-based multiplier</li> </ul> <h2>Last Value + Rule-based Multiplier</h2> <p>As everyone knows, this was a quite strong baseline. I wasn't able to beat this with other models so I decided to improve it by making data-driven decisions.</p> <p>First thing I did was adjusting last values to new year which is mentioned <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/389215" target="_blank">here</a> by <a href="https://www.kaggle.com/cdeotte" target="_blank">@cdeotte</a>. I got similar validation score boost when I tested this adjustment on my last three folds that were simulating new year transitions.</p> <p>The other thing I did was checking last 5 target values of each county:</p> <ul> <li>If target increased every time: Last value is multiplied with 1.010 for March, 1.015 for April and 1.015 May</li> <li>If target decreased every time: Last value is multiplied with 0.998 for March, 0.994 for April and 0.990 May</li> <li>If target was making sideways (else case): Last value is multiplied with 1.010 for March, April and May</li> </ul> <p>Lots of time passed since submissions are closed so I don't remember my exact validation scores but I remember that I got a significant boost with this approach. I didn't use Kaggle notebooks on this competition either. I was generating submissions on local and uploading them. This repository is the only thing I can share…<br> <a href="https://github.com/gunesevitan/godaddy-microbusiness-density-forecasting/" target="_blank">https://github.com/gunesevitan/godaddy-microbusiness-density-forecasting/</a></p> <p>and this is the script that generates 67th place predictions.<br> <a href="https://github.com/gunesevitan/godaddy-microbusiness-density-forecasting/blob/main/src/county_models/last_value.py" target="_blank">https://github.com/gunesevitan/godaddy-microbusiness-density-forecasting/blob/main/src/county_models/last_value.py</a></p>
GoDaddy - Microbusiness Density Forecasting	6th Place Solution - Lightgbm with Target flattened	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Congratulations winning the 1st place!</p> <p>This formula is construct by testing without theoretical basis. Here is the procedure:</p> <ol> <li>get the initial target:<br> <code>raw['target'] = raw.groupby('cfips')['microbusiness_density'].shift(-1)</code><br> <code>raw['target'] = raw['target']/raw['microbusiness_density'] - 1</code></li> <li>plot the target distribution for different active base.The distribution of target variables changes with the change of active base. The lower the active base is, the higher the absolute value of the target is.<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F425081%2F2d5b2ff154d7bcf3b275d00c216cc819%2FWechatIMG1609.png?generation=1687007781010306&alt=media" alt=""><br> 3.Assume that after transforming all active base into same target distribution,the distribution will be flattened like a straight line.Then find a suitable transform formula just like this:<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F425081%2F6a3bada85896f94a78b9c9d3eb36b525%2FWechatIMG1608.png?generation=1687007922643410&alt=media" alt=""></li> </ol> <p>Great Thanks to kaggle and Godaddy to hold this competition. Congratulations to all the winners. And many thanks go to people who created some mazing public notebooks.<br> Here is my Solution:</p> <h1>Validation</h1> <ul> <li>Last 12 month data for validation. </li> <li>To evaluate any solution is validate, I additionally compute the number of improve months of the 12 valid month instead of simple average cv improve.Because some month have abnormal high/low raising rate:<code>mbd(curr_month+1)/mbd(curr_month)-1</code>,average cv may lead to overfit</li> </ul> <h1>Outlier smooth</h1> <ul> <li>Mean smoothing for 2022-06->2022-08 to handle the general sharp rise-fall phenomenon <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F425081%2F97e293203e0ddc5a997612331e424658%2F2023-06-17%2017.54.47.png?generation=1686995741884356&alt=media" alt=""><ul> <li>There are also other data points have the phenomenon like "origin->Sharp rise-> Sharp fall->origin".But I haven't found the best way to deal with them yet.</li></ul></li> <li>Basic constant smoothing by fraction from giba</li> <li>Stable blacklist for those nearly unchanged cfips<ul> <li>[28055, 13101, 13265, 31009, 31115, 31149, 38047, 38087, 48033, 48301]</li></ul></li> </ul> <h1>features：</h1> <ul> <li>Action diff, Target shift, Target diff</li> <li>Target diff window <ul> <li>Sum</li> <li>Std</li></ul></li> <li>Target window<ul> <li>Sum</li> <li>Quantile 0.2</li> <li>Quantile 0.8</li> <li>Lowwer bound：<ul> <li>only have value when sign(Quantile 0.2)=sign(Quantile 0.8)，0 for other situations</li> <li>Quantile 0.8 for negative value</li> <li>Quantile 0.2 for positive value</li></ul></li></ul></li> <li>State cluster</li> </ul> <h1>Enhance modeling framework</h1> <p>construct the Enhance model frame, input 1)feature and 2) n_cross: number of future month to predict, output the raising rate of mbd in future n_cross month. In this way we can directly predict the public/private mbd by changing n_cross from 1-5<br> Detail structure:</p> <ul> <li>Enhance modeling frame features=features+n_corss</li> <li>n_corss=num of month to predict future mbd</li> <li>output target：[mbd(curr_month+n_corss)/mbd(curr_month)]*(1/n_corss)-1</li> <li>Model：Xgboost </li> </ul> <h1>Target_flatten：Train with flattened target, and do reverse transform in prediction</h1> <ul> <li>The distribution of target variables changes with the change of active base. The lower the active base is, the higher the absolute value of the target is.<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F425081%2Fdac497d7f6b9266fcd2e81242c29cf3c%2FWechatIMG1604.jpeg?generation=1686995149378846&alt=media" alt=""></li> <li>Do a simulate to transform all active base into same target distribution. We assume the distribution will be flatttened after the transformation, in this way we can construct the transform formula by testing:<ul> <li><code>coef=(1/(0.007(active_series/10+105))+1)</code></li> <li><code>target_flatten=target_series/coef</code></li></ul></li> <li>Train with flattened target, and do reverse transform in prediction</li> </ul> <h1>Post process</h1> <ul> <li>I noticed that the trend of public leaderboard month is abnormal higher than normal months.This phenomenon also happends on other month like 2022-07, In my opnion, such whole month trend is unpredictable, our model should not overfit it.</li> <li>Considering that the mbd for the public month is the basis for the subsequent private month, I post-processed the forecast results like:<br> <code>prediction=prediction*(1+0.001)</code></li> </ul>
GoDaddy - Microbusiness Density Forecasting	99th Place Solution - LightGBM and Rule-based Multiplier	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h1>Basic strategy</h1> <h3>How to get 5 months predictions</h3> <p>In this competition, we were required to make predictions for up to 5 months ahead. I decided to obtain 5 months of predictions by repeating the 1 month forecast five times. This approach made it easier to take advantage of the excellent predictions for January 2023 from the Public Notebook.</p> <h3>Validation</h3> <p>For validation, I used 5 months of data from August 2023 to December 2023. For the 1-month ahead forecast, I used data up to November 2023 for the December 2023 prediction and data up to July 2023 for the August 2023 prediction. For the 5-month ahead forecast, I used data up to July 2023 for the December 2023 prediction and data up to March 2023 for the August 2023 prediction. Using this method, I created predictions for each month as 1-month ahead to 5-months ahead predictions. I adjusted the models and features based on this validation score.</p> <h3>External data and Population Adjustment</h3> <p>Based on <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/389215" target="_blank">Chris Deotte's explanation</a>, I adjusted the microbusiness density values based on population, ensuring that the denominator of microbusiness density was the population of the same year throughout the data. As external data, I incorporated the census data given by the host and the data of cfips locations into the model. I did not utilize any other external data because they didn't improve my validation scores.</p> <h1>Models</h1> <h3>Overview</h3> <p>The overview of the prediction model is as follows. I used the model of <a href="https://www.kaggle.com/code/titericz/better-xgb-baseline" target="_blank">Giba's excellent notebook</a> as the starting model. Then I made rule-based adjustments to get the prediction values. I got hints for rule-based adjustments from Chris Deotte's notebook (<a href="https://www.kaggle.com/code/cdeotte/linear-regression-baseline-lb-1-092" target="_blank">here</a> and <a href="https://www.kaggle.com/code/cdeotte/seasonal-model-with-validation-lb-1-091" target="_blank">here</a>).</p> <h3>Base model</h3> <p>For the base model, I mostly adopted the preprocessing (including outlier corrections), feature engineering and target transformation from Giba's notebook. I used a LightGBM model with a small number of trees (n_estimators=30) as the model since complex models tended to overfit quickly according to the validation results. The remaining settings were determined through experiments based on the validation score. </p> <ul> <li>I didn't use a blacklist. </li> <li>I tried several settings for outlier correction and the number of lag and window variables</li> <li>I excluded cfips with less than 140 active values from models.</li> </ul> <p>This model gives an upward trend for most cfips. As pointed out in various notebooks, microbusiness density showed a global upward trend, so I believe this model roughly captures that trend.</p> <h3>Adhoc rules</h3> <p>I implemented two types of rule-based adjustments: trend-based and seasonal effects. These adjustments exclude cfips with low active values. For trend-based adjustments, I prepared rules such as "For cfips that have increased 1+x times in the last n months (or n-1 months), the average of the growth rate is multiplied by 1-y to obtain the new growth rate." Since applying the same rule repeatedly up to 5 months ahead often leads to overfitting, my adjustments become more conservative for farther predictions. For example, when creating the May 2023 prediction from the April 2023 prediction, trend-based adjustments are hardly applied.</p> <p>Regarding seasonal effects, I assumed effects based on months and prepared rules such as "For cfips that have increased by 1+x times or more compared to the previous month every time in the same month in the past, the average of the growth rate is multiplied by 1-y to obtain the new growth rate." It might be difficult to obtain conclusive evidence of the monthly effects based on the training data alone, but considering that what we want to predict is the economic index and validation scores were improved by this rule , I assumed their existence. </p> <p>In these rule-based adjustments, I implemented both upward and downward adjustments. However, considering the property of SMAPE, where downward deviations are less desirable than upward deviations, I made aggressive adjustments in the upward direction and conservative adjustments in the downward direction. In case of overlapping adjustments, I took the average. The threshold and detailed settings of the rules were determined through experiments based on the validation score.</p> <h1>Getting predictions and creating submission files</h1> <p>To get predictions, I used the best score from the Public Notebook as the starting point for the January 2023 prediction and incorporated the results of Public LB probing. The Public LB probing consists in the same rule-based approach described above. I applied multiple adjustments for both long-term and short-term trends in addition to seasonal effects. Using the obtained January 2023 prediction as the starting point, I made predictions up to May 2023 by repeating the 1-month ahead forecast. I also applied adjustments shown in <a href="https://www.kaggle.com/code/vitalykudelya/use-discrete-nature-of-microbusiness-density" target="_blank">Vitaly Kudelya's notebook</a> to the obtained predictions.<br> We can submit two files as final submission. For one submission, I submitted the prediction where rule-based adjustments were aggressive, and for the other submission I submitted the prediction where rule-based adjustments were modest. As a result, the former one was better than the latter one, which held also in validation scores.</p> <h1>Acknowledgements</h1> <p>Most of the basic ideas were fairly simple and derived from public notebooks. Setting up appropriate validations was cumbersome, but by attempting the combinations of these ideas based on the validation scores, I was able to achieve this rank. Thanks to Kagglers for many thoughtful and helpful notebooks and discussions. In competitions like this one, where the Public LB and Private LB data differ significantly, it can be difficult to determine what to trust, but preparing reliable validation is crucial above all else. My code is available at <a href="https://github.com/KazuakiTokunaga/kaggle-godaddy" target="_blank">github</a>. </p>
GoDaddy - Microbusiness Density Forecasting	??? Place Solution - Clustering County Stats	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>I would train on data up until the time of prediction so that I was not leaking "future" data. For example, the first validation window looks like the following. I have external features up until 2022-07-01, but am predicting the change in microbusiness density from 2022-04-01 to 2022-09-01.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5570735%2F419b7607949a90a2e2c4d48fe5837cdc%2FUntitled%20Diagram%20(4).jpg?generation=1678923210797113&alt=media" alt="Drawing"></p> <p>For the 2-step model I only validated on the final 4 months. The prediction windows were: </p> <p>2022-07-01 -> 2022-09-01 (last known mbd: 2022-04-01)<br> 2022-08-01 -> 2022-10-01 (last known mbd: 2022-05-01)<br> 2022-09-01 -> 2022-11-01 (last known mbd: 2022-06-01)<br> 2022-10-01 -> 2022-12-01 (last known mbd: 2022-07-01)</p> <p>First of all, thank you to GoDaddy and <a href="https://www.kaggle.com/kellenjgracey" target="_blank">@kellenjgracey</a> for hosting this awesome competition. Also, thank you to everyone who shared ideas in the discussion forums.</p> <p>The main idea of my solution was to fit 3x6x3 XGB models. I will try my best to explain this below.</p> <p><strong>Clustering</strong></p> <p>The idea was as follows. </p> <ol> <li>Order the data</li> <li>Split the data into 3 equal size groups</li> <li>Fit 3 XGB models (9 total) to each group (0-step, 1-step, and 2–step ahead)</li> <li>For each cfips, if predicting the last value outperformed XGB for all step-ahead models then assign that cfips to a 4th cluster.</li> </ol> <p>I performed the steps above with 6 features</p> <ul> <li># people age 30-34</li> <li># people w/ a tablet or other portable wireless computer</li> <li>PCT of population w/ two or more races</li> <li>County GDP</li> <li>PCT of businesses w/ between 10-19 employees</li> <li>Price reduction in homes over the past 12 months</li> </ul> <p>And the size of each cluster was as follows (the 4th value is the last-value cluster).</p> <ul> <li>[952, 933, 963, 287]</li> <li>[951, 948, 956, 280] </li> <li>[945, 945, 963, 282]</li> <li>[949, 951, 933, 302]</li> <li>[943, 948, 969, 275]</li> <li>[951, 937, 950, 297]</li> </ul> <p><strong>Data/Features</strong></p> <p>I then collected daily, weekly, and monthly features from <a href="https://fred.stlouisfed.org/" target="_blank">Fred</a>, and used data up until the 14th of each month (the day the competition would end). Some of these features came from the <a href="https://fred.stlouisfed.org/release?rid=10" target="_blank">Consumer Price Index Data</a> and <a href="https://fred.stlouisfed.org/release?rid=454" target="_blank">Average Price Data</a> which were released on the morning of the competition deadline.</p> <p><strong>Thoughts</strong></p> <p>My biggest concern here is data leakage. I tried to make sure that each feature was “known at the time of prediction”, and I hope I caught all data leaks when calculating my CV. (2nd biggest concern is overfitting economic indicators)</p> <p><strong>CV Score</strong></p> <p>The average SMAPE scores I calculated on the last 5 months of 2022 for each cluster are as follows. The ENS score takes an average of the 6 model predictions. </p> <table> <thead> <tr> <th></th> <th>Last</th> <th>Ens</th> <th>_008E</th> <th>_007E</th> <th>_049N</th> <th>dp_pc</th> <th>9_pct</th> <th>c_max</th> </tr> </thead> <tbody> <tr> <td>0-step</td> <td>2.8514</td> <td><strong>2.5820</strong></td> <td>2.6108</td> <td>2.6638</td> <td>2.6660</td> <td>2.6590</td> <td>2.6506</td> <td>2.6436</td> </tr> <tr> <td>1-step</td> <td>3.4091</td> <td><strong>2.6355</strong></td> <td>2.7057</td> <td>2.6922</td> <td>2.7026</td> <td>2.7047</td> <td>2.7245</td> <td>2.7082</td> </tr> <tr> <td>2-step</td> <td>4.1169</td> <td><strong>2.8772</strong></td> <td>2.9679</td> <td>2.9728</td> <td>2.9056</td> <td>2.9673</td> <td>2.9920</td> <td>2.9623</td> </tr> <tr> <td>Total</td> <td>3.4592</td> <td><strong>2.6982</strong></td> <td>2.7615</td> <td>2.7762</td> <td>2.7580</td> <td>2.7770</td> <td>2.7890</td> <td>2.7714</td> </tr> </tbody> </table>
GoDaddy - Microbusiness Density Forecasting	ETS, Linear Models, and MARS	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Below is a brief description + thoughts on my forecasting solution.<br> <strong>Private LB Place</strong>: 122/3,546<br> <strong>Private LB Score</strong>: 3.9884<br> <strong>Code</strong>: <a href="https://www.kaggle.com/code/abandura/mbd-forecasting-w-mars-lms-r" target="_blank">https://www.kaggle.com/code/abandura/mbd-forecasting-w-mars-lms-r</a></p> <p>Thanks to everyone who shared their insights and code!</p> <h2>Pre-Processing</h2> <p>Many counties exhibit level shifts in their historical microbusiness density, which reduces overall accuracy if simply included in the modeling process. I separated these from the rest of the counties by identifying any counties with a month-to-month change greater than 25%, and simply used last value predictions for them. Many other contestants took a similar approach. Looking at the size distribution of these counties (visualized in the histogram below) we can see that it excludes most of the smallest counties as well as a group of larger volume counties.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13474157%2F12985b7771d35940abaa8a6db8e72484%2Fexcl_cfips.png?generation=1687123492757230&alt=media" alt=""></p> <p>Small counties are unpredictable, as a single domain registration can dramatically change the density. However, looking at the larger counties with high variance it's clear that there is still a strong signal in many of them that could be leveraged for better predictions than a simple last value. For example, county ID 24013 has only one level shift up early on in the data set, and data before that shift could be removed and then interpolated so that it could be included in the forecast model training/prediction process. County IDs 10003 and 32510 have two level shifts that creates a "hump", which could be corrected in a similar manner using interpolation. I had planned to implement an algorithm to correct shifts for these larger counties, but unfortunately ran out of time. </p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13474157%2F6b3c0541a03e4f405b1aa745b943d5b6%2Flarge_excl_cfips.png?generation=1687123510916081&alt=media" alt=""></p> <p>Like many others, I also made the "percent change" adjustment to microbusiness density before training so that loss metrics like RMSE would weight small counties higher in the same way that SMAPE does. I used that adjusted target for all training other than the linear models, because they can be flexibly trained to directly minimize SMAPE (as described below).</p> <h2>Traditional Time-Series Forecasting, and other features</h2> <p>Of the traditional autoregressive time-series forecasting methods (ARIMA, Holt-Winters, Exponential Smoothing, etc.) that I tried, Exponential Smoothing (ETS) was the best. I ended up using ETS forecasts as features for the linear modeling by rolling through the data and generating ets forecasts for 1-6 month time horizons.</p> <p>Other features I toyed with included moving-average terms, difference terms, month and year, and the average of all previous MB density values. </p> <h2>CV Strategy</h2> <p>For CV I used the last 6 months of data, as shown below, and trained separate models for each forecast horizon. I wanted to use as much data as possible for training, since we only had a few years of historical data spanning the pandemic.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13474157%2F6ae7279f1d78b31a125355f0c2d34cd8%2Fcv_strategy.png?generation=1687123527511586&alt=media" alt=""></p> <h2>Linear Modeling via Optimization</h2> <p>I tried other ML approaches like XGBoost, but found the most success with the simpler linear models.</p> <p>Another nice feature of linear models is you can directly fit them to minimize SMAPE by solving the optimization problem with the following objective:</p> <pre><code> fn par df r_cols y norm_col l2 l1 preds par df.N l2p l1p ind par i_col r_colsind preds preds parinddf geti_colmeangeti_colsdgeti_col l2p l2p l2parind l1p l1p l1parind err smapeadfgety fcpreds l1p l2p err </code></pre> <p>I tested L1 and L2 regularization, but found the lowest CV accuracy with no regularization. Looking at the fitted weight magnitudes, it seems that MA terms are most important in the forecasts. This was my best-scoring final submission.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13474157%2F8b4a56ee77fff6fe87562ab6916726f5%2Flm_weights.png?generation=1687130521279924&alt=media" alt=""></p> <h2>Multivariate adaptive regression spline (MARS)</h2> <p><a href="https://en.wikipedia.org/wiki/Multivariate_adaptive_regression_spline" target="_blank">https://en.wikipedia.org/wiki/Multivariate_adaptive_regression_spline</a></p> <p>Because I had success with the LMs, I though that MARS would be a natural improvement since it is an algorithm for fitting piecewise linear models to automatically deal with non-linearities and incorporate variable interactions.</p> <p>I did some hyperparameter tuning, but had the most success with the default parameters, and based on the CV accuracy I hoped it would outperform the LMs. However that wasn't the case with the private LB (I'm still not sure why, perhaps MARS was more sensitive to erroneous Dec data…)</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13474157%2Fb41c10f7948ea74940de90bd4e8f357a%2Fsmape_comp.png?generation=1687127830752749&alt=media" alt=""></p>
GoDaddy - Microbusiness Density Forecasting	Giba RAPIDS SVR Solution	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>For my first solution I used a global linear multiplier were I pick the coefficient optimizing SMAPE for each split month and forecast span separately.</p> <p>The second solution is based in SVR. You can check my notebook here: <a href="https://www.kaggle.com/code/titericz/super-fast-rapids-svr/notebook" target="_blank">Super Fast RAPIDS SVR</a></p> <p>Once <code>microbusiness_density</code> can be calculated using <code>active</code> and <code>county population</code>, I used <code>active</code> as the target labels in my models.<br> For validation I used last 12 months to calculate the SMAPE for each forecast range. To ease development I tracked only the gain each algorithm gives compared with last value benchmark. For example, predicting 3 months ahead with last value gives SMAPE 2.717, if my algorithm scores 2.617, then the gain will be 0.10 over the baseline.</p> <p>The SVR gain for each forecast period is:</p> <table> <thead> <tr> <th>Forecast</th> <th>SMAPE Gain(12 months avg)</th> </tr> </thead> <tbody> <tr> <td>1</td> <td>0.0171</td> </tr> <tr> <td>2</td> <td>0.0611</td> </tr> <tr> <td>3</td> <td>0.1555</td> </tr> <tr> <td>4</td> <td>0.2621</td> </tr> <tr> <td>5</td> <td>0.3524</td> </tr> <tr> <td>6</td> <td>0.4750</td> </tr> </tbody> </table> <p>As you can see the gain is not much. But all other algorithms I tried (including GBDTs) scored worse than SVR in my approach.</p> <h1>Other than that I wish good luck to everyone that dedicated time to this competition! 😉</h1> <p>obs. </p> <ul> <li>notebook V1 using sklearn SVR in 4607s.</li> <li>notebook V2 using RAPIDS SVR 136s. (34x faster 💪)</li> </ul> <p>SVR around ~20-30% better than XGB/LGB. </p>
GoDaddy - Microbusiness Density Forecasting	Godday's Solution that didn't Appear in Public Notebooks.	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thank you that you like it. If my solution works well in Private LB, I will share my code.</p> <p>Yes, it is quite strange. However, my CV result on 5th month is 3.15, which is more reasonable and reliable.</p> <p>Thank you. I think your work gets a much better CV score than mine, congrats. <a href="https://www.kaggle.com/tarique7" target="_blank">@tarique7</a> </p> <p>This competition is full of positions, and I don't know if I can get a good result in the end. But I figured I'd forget about the game if I waited another 3 months, so I decided to share my solution anyway.</p> <h2>0. Preface:</h2> <p>Hi all people diving in this competition,</p> <p>We had a very memorable racing experience together (at least for me). I learned a lot from this competition, and to be honest, I'm just a newbie in machine learning and time series research. I met a lot of talent here <a href="https://www.kaggle.com/cdeotte" target="_blank">@cdeotte</a> <a href="https://www.kaggle.com/kaggleqrdl" target="_blank">@kaggleqrdl</a> <a href="https://www.kaggle.com/joshuascj" target="_blank">@joshuascj</a> <a href="https://www.kaggle.com/titericz" target="_blank">@titericz</a> <a href="https://www.kaggle.com/batprem" target="_blank">@batprem</a> <a href="https://www.kaggle.com/vitalykudelya" target="_blank">@vitalykudelya</a> . Too many friends for me to put them all here.<br> They kindly shared good baselines, ideas and data, which taught me a lot. I must say that we have reached such score by standing on the shoulders of giants.</p> <h2>1. solution:</h2> <h3>1.1. ideas</h3> <p>Our team used GRU blending with XGBoost as a solution.<br> The composition of time series data is probably: (this is my understanding, I hope there is no error in expression)</p> <p>$$timeseries=trend+otherinformantion+noise$$</p> <p>The general idea is to use GRU to capture the "trend" information of the data, and XGBoost to capture "other information".</p> <p>Expressed in pseudocode:</p> <p><strong>Fit stage</strong></p> <p><code>GRU. fit(trend_data)</code><br> <code>XGB.train(X_XGB[other_information_features], residual_data)</code></p> <p><strong>Inference stage</strong><br> <code>residual_data_pred=XGB.predict(X_XGB[other_information_features])</code><br> <code>trend_data_pred = GRU. predict(X_GRU)</code><br> <code>sub_data_pred=trend_data_pred +residual_data_pred</code></p> <h3>1.2. Cross-validation</h3> <ol> <li>When we train GRU, I predict 5 months totally. I use rolling-window CV method for 5 times.</li> </ol> <p><strong>Results:</strong> 1st month: 1.333; 3rd month: 2.47; 4th month: 2.95; 5th month: 3.39</p> <ol> <li>When we train XGB, we use Group CV results for 3 times.<br> <strong>Results:</strong> 1st month: Nan; 3rd month: 2.34; 4th month: 2.74; 5th month: 3.12</li> <li>Improve:</li> <li><strong>Results:</strong> 1st month: Nan; 3rd month: 0.13; 4th month: 0.20; 5th month: 0.26</li> </ol> <h3>1.3. LB tests:</h3> <p>Also, I use LB to determine my CV’s performance.<br> I treat January (LB) as "March", "April" and "May" respectively to output the results. (Just slide the GRU forecast starting point backwards from 2022-12-01.)</p> <p><strong>Only GRU results:</strong> 3rd month: 2.82, 4th month: 3.15, 5th month: 3.4</p> <p><strong>GRU blending XGB</strong>: 3rd month: 2.66, 4th month:2.8, 5th month: 2.5</p> <p><strong>Improve</strong>: 3rd month: 0.16, 4th month:0.35, 5th month: 0.9 (Maybe this is somehow the luck)</p> <h3>1.4. Conclusion</h3> <p>Both CV score and LB score all get improvement. XGB can help GRU. Thank you. <br> Good luck to us.</p> <p>Trends do not exist in every county, but most counties with high active/popularity have trends, (up or down). But there are a small number of (very few) towns with very low activity that also have a relatively stable upward trend. </p> <p>My model tells me that the overall/global trend is all up, but slightly. Somehow it is similar to the Linear Model results, that's why linear model works well in this competition I guess.</p> <p>As analyzed by many Kagglers, last value is a good choice for small counties, because small county didn't have a stable trend. Instead of guessing whether they will rise or fall in the future, it is better to keep the current value unchanged and you may get more Good SMAPE.</p> <p>Thank you for your comment! <a href="https://www.kaggle.com/cpmpml" target="_blank">@cpmpml</a> Yes, my solution is quite similar to yours, because I did browse your code before I implement my solution. It did work to improve GRU.</p> <p>thank you for your good wishes! <a href="https://www.kaggle.com/siddharthkumarsah" target="_blank">@siddharthkumarsah</a> . I hope my solution can get a satisfied result, but there are so many talents in this competition. Anyway, wish us good luck.</p>
GoDaddy - Microbusiness Density Forecasting	My Forecasting Strategy	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h1>Trend</h1> <p>My cross-validation didn't show that complicated gradient boostings showed significantly better results than simple approaches. Also for some months gradient boostings performed worse than the Last Value baseline and simple approaches showed more stable performance. </p> <p>The following formula shows how I estimated trend for cfips.<br> <code>active_quantile_group_21_last_6m_trend</code> means the best multiplication constant (clipped from 1 to 1.006) for the last value of <code>microbusiness_density</code> for the next month on a previous 6 month for a group of cfips (total 21 groups by quantile of active). The same logic for <code>state</code> and concrete <code>cfips</code> group.<br> <code>forward_2</code> means the best multiplication constant for the next 2 month (clipped from 1 to 1.012).</p> <pre><code>df_features_data[] = ( df_features_data[] * + df_features_data[] * + df_features_data[] * + df_features_data[] * + df_features_data[] * + df_features_data[] * ) * + (( df_features_data[] * + df_features_data[] * + df_features_data[] * )*) </code></pre> <h1>Using the best public submission as a starting point for predictions</h1> <p>Public Leaderboard data wasn't published and for time-series data it should be beneficial to use the best public submission.</p> <h1>Select 2 Submissions. Positive and Negative</h1> <p>It's a good idea to cover different scenarios selecting 2 submissions. I call it Positive submission (we under-forecasting trend) and negative (we over-forecasting trend). <br> For example for Positive submission <code>* (trend + 0.0005)*(month_number)</code></p> <h1>May Seasonality</h1> <p>I just used forecast for April <code> (1 - 0.0025)</code> for Positive submission, because May showed a dropdown of values the last two years.</p> <h1>Cfips with active < 150</h1> <p>My cross-validation showed that using the last value baseline for such cfips (as many public baselines do) is good only for one month forward forecast, for March, April, May i used some multiplication constants.<br> The following plot shows that for active < 75 using the last value baseline for 3 month forecast is the best, for 75-150 it's better to use trend. <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F297456%2F7059d3c2b37887d58d0ae4b88dd4cb4d%2Fnewplot%20(98).png?generation=1678841088819412&alt=media" alt=""></p> <h1>Hope that random helps 🤞</h1>
GoDaddy - Microbusiness Density Forecasting	My Strategy & The Dangers of Public Notebooks	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Yes, this is a really good point I didn't bring up in my original post and something that I was careful to do. Once I predicted a new value, the main features I had to update for my notebook were the lag features. Since I used lagged active features too, this also required me to compute the active feature for each predicted time step from the predicted value and the 2020 population (later adjusted with the 2021 population) via the following code:</p> <pre><code>active_mask = (raw["dcount"] == (TS + 1)) raw.loc[active_mask, "active"] = ((raw.loc[active_mask, "pop2020"] * raw.loc[active_mask, "ypred"]) / 100).astype(int) </code></pre> <p>Then updating the lag features with a function I made: <code>raw = update_lags(raw)</code></p> <p>Thanks for your comment! </p> <p>I'm not 100% sure what you mean by reusing the model many times. If I understand correctly, "reusing the last value model many times, and generating several predictions" is indeed what I did and what I meant when I said "autoregressively" generate predictions. Like most models, this likely has it's upsides and downsides. From my analysis, there seems to be a left-skew in the distribution of my predicted values (thus my predictions will tend to be more positive than negative). As an alternative method, it seems other people like <a href="https://www.kaggle.com/joshuascj" target="_blank">@joshuascj</a> used models trained with "gaps" to predict the following months, find his explanation <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/394822" target="_blank">here</a>. </p> <p>Hello everyone,</p> <p>I may not have the best performing model or the most clever, but I thought I would share my strategy for people interested. Prior to this competition I had no time series experience and learned a ton during this competition, so thanks to everyone who made their notebooks public and shared their ideas. My strategy consists of two different strategies/models. </p> <p>Things I tried that did not work: GRU's, linear regression models (I had to be too selective with counties to get good performance), and traditional autoregressive models like ARMA didn't work for me. If anyone got these to work well, I'de love to hear how you did it. </p> <p><strong>First model:</strong> My first model is just a “last value” model and is meant to be a hedge against my main model. The strategy was this: Overfit the leaderboard to leak back the best values to use for a “last value” model. I figure since most public notebooks work via predicting the first value of the public lb and just repeating that value, then I should be in a pretty decent place if I can outperform the top public notebooks in the public LB (although there are no guarantees this will outperform in private LB).</p> <p><strong>Second model:</strong> My second model was a modified version of this notebook:</p> <ul> <li><a href="https://www.kaggle.com/code/vadimkamaev/micro-business-catboost" target="_blank">Micro-Business. Catboost</a>. </li> </ul> <p>For my modifications I did the following:</p> <ol> <li><p>First I performed a custom preprocessing scheme that I built, followed by another preprocessing scheme from the popular <a href="https://www.kaggle.com/code/titericz/better-xgb-baseline" target="_blank">Better XGB Baseline</a> notebook. More information on my preprocessing scheme can be found in my notebook explaining it here: <a href="https://www.kaggle.com/competitions/godaddy-microbusiness-density-forecasting/discussion/393364" target="_blank">Outliers, Changepoints, and Smoothing</a></p></li> <li><p>Next, rather than only predicting the first value, I modified the code to autoregressively predict each value in the testing set. </p></li> <li><p>I incorporating the ideas in the following notebook: <a href="https://www.kaggle.com/code/vitalykudelya/use-discrete-nature-of-microbusiness-density" target="_blank">Use Discrete Nature of microbusiness_density</a></p></li> <li><p>I removed the meta model and just did a threefold ensemble of an XGBoost, CatBoost, and LGBM model. </p></li> <li><p>Tons of other "micro-adjustments" to preprocessing, feature engineering, training, and postprocessing.</p></li> <li><p>Removed "suspicious" looking code. See the topic below.</p></li> </ol> <p><strong>On the Dangers of Public Notebooks:</strong></p> <p>It seems that the five best scoring public notebooks (maybe more, I stopped looking after five) have a fatal flaw in their submissions resulting from the following code at the bottom of each notebook:</p> <pre><code>for i, row in sub.iterrows(): test.iat[i,1] = ( 0.65test.iat[i,1] + 0.35row["microbusiness_density"] ) test.to_csv('submission.csv', index=False) test.head() </code></pre> <p>This block of code will essentially cause all the “private LB” values to be 0.35 times the last predicted public LB value. Therefore, if users who blindly copied the top public notebooks didn’t check for this, their models will likely not generalize to the private LB data. Whether this was done intentionally or not, I do not know. In total, these notebooks had 2545 copies and edits (likely from a lot of the same people copying). Below is a sample from these submissions.</p> <table> <thead> <tr> <th>row_id</th> <th>microbusiness_density</th> </tr> </thead> <tbody> <tr> <td>1001_2022-11-01</td> <td>3.404743687834736</td> </tr> <tr> <td>1001_2022-12-01</td> <td>3.422746298213241</td> </tr> <tr> <td>1001_2023-01-01</td> <td>3.3372338989153425</td> </tr> <tr> <td>1001_2023-02-01</td> <td>1.1679193483055044</td> </tr> <tr> <td>1001_2023-03-01</td> <td>1.1679193483055044</td> </tr> <tr> <td>1001_2023-04-01</td> <td>1.1679193483055044</td> </tr> <tr> <td>1001_2023-05-01</td> <td>1.1679193483055044</td> </tr> <tr> <td>1001_2023-06-01</td> <td>1.1679193483055044</td> </tr> </tbody> </table> <p>Best of luck in the coming months!</p> <p>Ryan P.</p>
GoDaddy - Microbusiness Density Forecasting	Public 1125 - > Private 48 solution	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Methods that are beneficial for prediction:</p> <p>Outlier smoothing<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5998972%2Fbc31d3c4fcf04c23e7afb091f931984b%2F1.png?generation=1687061025743863&alt=media" alt=""><br> Coefficient adjustment + approximation<br> active: The raw count of microbusinesses in the county The target to be predicted is population density = active / population. The population statistics change every January. The competition requires predicting the population density in 2023, which has changed in the denominator of population density. Based on the existing data, use (2023 predicted value * 2020 population) / 2021 population to estimate the coefficient of population trend change, and then correct the predicted value with this coefficient. Because the numerator part active is originally an integer, the predicted value multiplied by the population is usually a continuous value, such as 1.3, while the actual value may be 1, and rounding can help increase accuracy. The predicted value after coefficient correction should be multiplied by the 2021 population, then rounded, and then divided by the 2021 population for further correction.</p> <p>Optimal shift time = 12<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5998972%2Fc78306ebaa11dbf3adbc2c6a1d6a1b62%2F2.png?generation=1687061119239898&alt=media" alt=""><br> Create lagged features for the past 12 months, if it is 24 or 6, it will make the effect worse, especially 24.</p> <p>The state and county fields are label encoded, plus the cfips field (equivalent to state + county), which are beneficial for the prediction results. <br> Rate of change effect is better<br> For predicting n + gap tasks, change the original density continuous value to be predicted to the growth value between the current value and the previous gap months.<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5998972%2F9c4f2fa8203de8a5e29cc538e1dfccb8%2F3.png?generation=1687061133256397&alt=media" alt=""><br> Make some lagged features for the original density continuous value, and then add some lagged features for this growth value, which improves the prediction accuracy</p> <p>After converting the density target value into a growth value, LightGbm and Xgboost models need to change the default objective parameter and use pseudo huber loss as the optimization indicator.</p> <p>========</p> <p>对预测有增益的方法：</p> <h1>异常值平滑</h1> <h1>系数调整 + 求近似值</h1> <p>active：The raw count of microbusinesses in the county<br> 要预测的目标为人口密度 = active / 人口。人口统计值每年一月会发生变化。竞赛要预测2023年的人口密度，在人口密度分母上已经发生了变化。基于已有的数据，使用（2023年预测值 * 2020年人口）/ 2021年人口估算出人口趋势变化的系数，再将该系数修正预测值。<br> 因为分子部分active本来是整数，预测值乘以人口通常是连续值，譬如1.3，而实际值可能是1，取整后可以帮助精度增加。系数修正后的预测值要乘以2021年人口，然后取整，再除以2021年人口，做进一步修正。</p> <h1>最优shift time = 12</h1> <p>创建过去12个月的滞后特征，如果是24或者6，都会让效果变差，尤其是24.</p> <h1>state和county两个字段做label encoding，加上cfips（等同于state + county）字段，都对预测结果有增益。</h1> <h1>rate of change 效果变好</h1> <p>1.预测n + gap的任务，将原本要预测的密度连续值，改成当前值与上gap个月的增长值。</p> <p>2.对原本对密度连续值做一些滞后特征，再加上对该增长值做滞后特征，对预测准度有提升<br> 3.将密度目标值转换成增长值后，LightGbm和Xgboost模型需要更改默认的objective参数，使用pseudo huber loss作为优化指标.</p>
GoDaddy - Microbusiness Density Forecasting	Public 531st -> Private 10th: The Complete Solution Code	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: GoDaddy - Microbusiness Density Forecasting <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h1><strong>GoDaddy - Microbusiness Density Forecasting</strong></h1> <p>The solution is 3 months ago, so if there is anything missing, please let me know.</p> <p>The objective of the competition is to predict monthly <strong>microbusiness density</strong> in <strong>3315 cfips</strong> (County FIPS) in the United States. The code implementation incorporates several special features and techniques to improve the accuracy of the predictions.</p> <h2><strong>Approach</strong></h2> <h3><strong>1. Predicting 3315 Time Series</strong></h3> <p>This approach predicts the microbusiness density for each of the <strong>3315 cfips</strong> individually. This allows for more granular and accurate predictions tailored to specific regions.</p> <h3><strong>2. Technical Indicators</strong></h3> <p>The code utilizes popular technical indicators such as <strong>Exponential Moving Average (EMA)</strong>, <strong>Momentum (MOM)</strong>, and <strong>Relative Strength Index (RSI)</strong>. These indicators capture underlying trends, momentum, and market conditions, enhancing the predictive power of the model.</p> <h3><strong>3. External Datasets</strong></h3> <p>In addition to the microbusiness density data, the model incorporates several external datasets. These datasets include information such as <strong>unemployment data</strong>, <strong>earnings</strong>, <strong>rent</strong>, <strong>DSG10</strong>, <strong>tax rate</strong>, <strong>housing price</strong>, and <strong>population estimates</strong>. By incorporating these relevant external factors, the model can capture the influence of broader economic and demographic factors on microbusiness density.</p> <h3><strong>4. Optuna for Hyperparameter Optimization</strong></h3> <p><strong>Optuna</strong> is used to minimize the <strong>SMAPE</strong> in the <strong>Catboost</strong> model.</p> <h3><strong>5. Catboost Model with Cross Validation</strong></h3> <p>The code employs the <strong>Catboost model</strong>. Also used <strong>Cross Validation (CV)</strong> to generalize well to unseen data.</p> <h3><strong>6. Multiple Model Training</strong></h3> <p>To predict multiple future time periods, five separate models are trained. Each model is designed to predict a specific time horizon, such as <strong>t+1 month</strong>, <strong>t+2 months</strong>, and so on.</p> <h3><strong>7. External Dataset Addition</strong></h3> <p>The code "<strong>6-external-datasets.ipynb</strong>" is found on Kaggle and is provided to incorporate additional external datasets into the prediction model.</p> <h2><strong>Limitations</strong></h2> <ul> <li><strong>Limited Historical Data</strong>: The current approach utilizes only the past three months of data to predict future microbusiness density. Exploring longer intervals of historical data could provide additional context and potentially improve the forecasting accuracy.</li> </ul> <h2><strong>Code Files</strong></h2> <ul> <li><p><strong>6-external-datasets.ipynb</strong>: add additional external datasets to the prediction model. It can be found on Kaggle and is intended to enhance the feature set.</p></li> <li><p><strong>kaggle_competition_microbusiness.ipynb</strong>: contains the full model training code for microbusiness density prediction. It encompasses data preprocessing, model training using <strong>Catboost</strong> and <strong>Optuna</strong>, and evaluation using <strong>SMAPE</strong>.</p></li> </ul> <p><strong>Github link</strong>: <a href="https://github.com/ttterence927/kaggle_competition_microbusiness/" target="_blank">https://github.com/ttterence927/kaggle_competition_microbusiness/</a></p> <h2><strong>Show Your Support</strong></h2> <p>If you find this code implementation valuable or interesting, please consider giving it a star on <strong>GitHub</strong>.</p> <p>Thank you :)) Learned a lot from you and all the people here. And congrats on winning 1st place!!</p>
Google - American Sign Language Fingerspelling Recognition	122nd Solution for Google - American Sign Language Fingerspelling Recognition Competition	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>We did this in our freshman summer, defintely learnt a lot.</p> <p><strong>Context</strong></p> <ul> <li>Business context:<a href="https://www.kaggle.com/competitions/asl-fingerspelling/overview" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/overview</a></li> <li>Data context:<a href="https://www.kaggle.com/competitions/asl-fingerspelling/data" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/data</a></li> </ul> <p><strong>Overview of the Approach</strong><br> Our work is basically some improvements based on the <a href="https://www.kaggle.com/code/royalacecat/the-deeper-the-better" target="_blank">public notebook baseline</a> w/ LB score 0.699, and we improved it w/ +0.007.<br> It conbines Transformer and 1D-CNN.<br> We basically improved it by doing some feature engineering to the frames.</p> <p><strong>Details on the Submission</strong><br> Originally our work was based on another public notebook baseline which only uses the Transformer, and we did some improvement on that by rescheduling the learning rate, specifically, to increase the learning rate at the end, but it didn't work for our final notebook.<br> We also processed the supplemental data, but it didn't have some improvements(maybe we did it the wrong way).<br> Our data augmentation includes some techniques like random cropping, rotation, affine transformations, etc.</p> <p><strong>Code Samples for Data Augmentation</strong></p> <pre><code>@tf.function() def : tf.[0] < FRAME_LEN: x = tf.pad(x, ([[, FRAME_LEN-tf.shape(x)[]], [, ], [, ]]), constant_values=()) : x = tf.image.resize(x, (FRAME_LEN, tf.shape(x)[])) x @tf.function() def resize_pad_channel(x): tf.shape(x)[] < FRAME_LEN: x = tf.pad(x, ([[, FRAME_LEN-tf.shape(x)[]], [, ], [, ]]), constant_values=()) : x = tf.image.resize(x, (FRAME_LEN, tf.shape(x)[])) x def interp1d_(x, target_len, method=): target_len = tf.maximum(,target_len) method == : tf.random.uniform(()) < : x = tf.image.resize(x, (target_len,),) : tf.random.uniform(()) < : x = tf.image.resize(x, (target_len,),) : x = tf.image.resize(x, (target_len,),) : x = tf.image.resize(x, (target_len,),method) resize_pad(x) def flip_lr(x): dx, dy, dz = tf.unstack(x, axis=) dx = -dx new_x = tf.([dx,dy,dz], ) new_x def resample(x, rate=(,)): rate = tf.random.uniform((), rate[], rate[]) length = tf.shape(x)[] new_size = tf.cast(ratetf.cast(length,tf.float32), tf.int32) new_x = tf.image.resize(x, (new_size, )) resize_pad_channel(new_x) def spatial_random_affine(xyz, scale = (,), shear = (,), shift = (,), degree = (,), ): center = tf.constant([,]) scale is not None: scale = tf.random.uniform((),scale) xyz = scalexyz shear is not None: xy = xyz[...,:] z = xyz[...,:] shear_x = shear_y = tf.random.uniform((),shear) tf.random.uniform(()) < : shear_x = : shear_y = shear_mat = tf.identity([ [,shear_x], [shear_y,] ]) xy = xy @ shear_mat center = center + [shear_y, shear_x] xyz = tf.concat([xy,z], axis=) degree is not None: xy = xyz[...,:] z = xyz[...,:] xy -= center degree = tf.random.uniform((),degree) radian = degree/np.pi c = tf.math.(radian) s = tf.math.(radian) rotate_mat = tf.identity([ [c,s], [-s, c], ]) xy = xy @ rotate_mat xy = xy + center xyz = tf.concat([xy,z], axis=) shift is not None: shift = tf.random.uniform((),shift) xyz = xyz + shift xyz def temporal_crop(x, length=FRAME_LEN): l = tf.shape(x)[] offset = tf.random.uniform((), , tf.clip_by_value(l-length,,length), dtype=tf.int32) x = x[offset:offset+length] x def temporal_mask0(x, size=(,), mask_value=()): l = tf.shape(x)[] mask_size = tf.random.uniform((), size) mask_size = tf.cast(tf.cast(l, tf.float32) * mask_size, tf.int32) mask_offset = tf.random.uniform((), , tf.clip_by_value(l-mask_size,,l), dtype=tf.int32) x = tf.tensor_scatter_nd_update(x,tf.range(mask_offset, mask_offset+mask_size)[...,None],tf.fill([mask_size,,],mask_value)) x def temporal_mask(x, rate=, mask_value=()): mask_size=(FRAME_LENrate) mask = tf.squeeze(tf.random.categorical(np.mat([/FRAME_LEN i in range(FRAME_LEN)]),mask_size)) x = tf.tensor_scatter_nd_update(x,mask[...,None],tf.fill([mask_size,,],mask_value)) x def spatial_mask(x, size=(,), mask_value=()): mask_offset_y = tf.random.uniform(()) mask_offset_x = tf.random.uniform(()) mask_size = tf.random.uniform((), size) mask_x = (mask_offset_x<x[...,]) & (x[...,] < mask_offset_x + mask_size) mask_y = (mask_offset_y<x[...,]) & (x[...,] < mask_offset_y + mask_size) mask = mask_x & mask_y x = tf.where(mask[...,None], mask_value, x) x def augment_fn(x, always=False): tf.random.uniform(())< or always: x = resample(x, (,)) tf.random.uniform(())< or always: x = flip_lr(x) tf.random.uniform(())< or always: x = spatial_random_affine(x) tf.random.uniform(())< or always: x = temporal_mask(x) tf.random.uniform(())< or always: x = spatial_mask(x) x @tf.function(jit_compile=True) def pre_process0(x): lip_x = tf.gather(x, LIP_IDX_X, axis=) lip_y = tf.gather(x, LIP_IDX_Y, axis=) lip_z = tf.gather(x, LIP_IDX_Z, axis=) rhand_x = tf.gather(x, RHAND_IDX_X, axis=) rhand_y = tf.gather(x, RHAND_IDX_Y, axis=) rhand_z = tf.gather(x, RHAND_IDX_Z, axis=) lhand_x = tf.gather(x, LHAND_IDX_X, axis=) lhand_y = tf.gather(x, LHAND_IDX_Y, axis=) lhand_z = tf.gather(x, LHAND_IDX_Z, axis=) rpose_x = tf.gather(x, RPOSE_IDX_X, axis=) rpose_y = tf.gather(x, RPOSE_IDX_Y, axis=) rpose_z = tf.gather(x, RPOSE_IDX_Z, axis=) lpose_x = tf.gather(x, LPOSE_IDX_X, axis=) lpose_y = tf.gather(x, LPOSE_IDX_Y, axis=) lpose_z = tf.gather(x, LPOSE_IDX_Z, axis=) lip = tf.concat([lip_x[..., tf.newaxis], lip_y[..., tf.newaxis], lip_z[..., tf.newaxis]], axis=) rhand = tf.concat([rhand_x[..., tf.newaxis], rhand_y[..., tf.newaxis], rhand_z[..., tf.newaxis]], axis=) lhand = tf.concat([lhand_x[..., tf.newaxis], lhand_y[..., tf.newaxis], lhand_z[..., tf.newaxis]], axis=) rpose = tf.concat([rpose_x[..., tf.newaxis], rpose_y[..., tf.newaxis], rpose_z[..., tf.newaxis]], axis=) lpose = tf.concat([lpose_x[..., tf.newaxis], lpose_y[..., tf.newaxis], lpose_z[..., tf.newaxis]], axis=) hand = tf.concat([rhand, lhand], axis=) hand = tf.where(tf.math.is_nan(hand), , hand) mask = tf.math.not_equal(tf.reduce_sum(hand, axis=[, ]), ) lip = lip[mask] rhand = rhand[mask] lhand = lhand[mask] rpose = rpose[mask] lpose = lpose[mask] lip, rhand,lhand, rpose, lpose @tf.function() def pre_process1(lip, rhand,lhand, rpose, lpose, augment = False): lip = (resize_pad(lip) - LIPM) / LIPS rhand = (resize_pad(rhand) - RHM) / RHS lhand = (resize_pad(lhand) - LHM) / LHS rpose = (resize_pad(rpose) - RPM) / RPS lpose = (resize_pad(lpose) - LPM) / LPS x = tf.concat([lip, rhand, lhand, rpose, lpose], axis=) x = tf.where(tf.math.is_nan(x), , x) augment: x = augment_fn(x) s = tf.shape(x) x = tf.reshape(x, (s[], s[]s[])) x pre0 = pre_process0(frames) pre1 = pre_process1(pre0) INPUT_SHAPE = (pre1.shape) print(INPUT_SHAPE) </code></pre> <p><strong>Some Useful References</strong><br> <a href="https://www.kaggle.com/code/royalacecat/the-deeper-the-better" target="_blank">https://www.kaggle.com/code/royalacecat/the-deeper-the-better</a><br> <a href="https://www.kaggle.com/code/hebasaleh00/aslfr-eda-preprocessing" target="_blank">https://www.kaggle.com/code/hebasaleh00/aslfr-eda-preprocessing</a><br> <a href="https://www.kaggle.com/competitions/asl-fingerspelling/discussion/409438" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/discussion/409438</a></p>
Google - American Sign Language Fingerspelling Recognition	126th Place Solution	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p><a href="https://www.kaggle.com/code/markwijkhuizen/aslfr-inference-python3-7-tpu" target="_blank">This solution</a> is an improved version of <a href="https://www.kaggle.com/code/markwijkhuizen/aslfr-transformer-training-inference" target="_blank">my public notebook</a> with the following modifications:</p> <ul> <li>Running on a TPU to allow for faster training and experimentation</li> <li>Increased number of input frames 128 → 288</li> <li>Batch size 64 → 512</li> <li>Units encoder 384 → 288</li> <li>Units decoder 256 → 128</li> </ul> <p>For an in depth explanation of the approach I refer to the public notebook code and comment section.</p> <p>The expect the performance boost in my private notebook is mostly due to the decreased model capacity and increased batch size, this should reduce overfitting.</p>
Google - American Sign Language Fingerspelling Recognition	17th Solution: Conformer + CTCLoss + 500 epoch training	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h1>TL; DR</h1> <p>Fixed-length input (220 frames), padding shorter inputs and resizing longer inputs.<br> 2 layer MLP landmark encoder + 6 layer 384-dim Conformer + 1 layer GRU and 500 epochs training (takes ~8 hours on Kaggle TPUs).<br> Post-processing (+0.003 in CV and LB) by:</p> <pre><code> len() <= : = + </code></pre> <p>My notebook is available at <a href="https://www.kaggle.com/code/nightsh4de/ctc-transformer/notebook" target="_blank">https://www.kaggle.com/code/nightsh4de/ctc-transformer/notebook</a><br> It is primarily based on the previous 1st solution <a href="https://www.kaggle.com/code/hoyso48/1st-place-solution-training" target="_blank">https://www.kaggle.com/code/hoyso48/1st-place-solution-training</a> and some public notebooks of this competition:<br> <a href="https://www.kaggle.com/code/markwijkhuizen/aslfr-transformer-training-inference#Landmark-Embedding" target="_blank">https://www.kaggle.com/code/markwijkhuizen/aslfr-transformer-training-inference#Landmark-Embedding</a><br> <a href="https://www.kaggle.com/code/irohith/aslfr-transformer" target="_blank">https://www.kaggle.com/code/irohith/aslfr-transformer</a> <br> <a href="https://www.kaggle.com/code/shlomoron/aslfr-ctc-on-tpu" target="_blank">https://www.kaggle.com/code/shlomoron/aslfr-ctc-on-tpu</a>. Many many thanks to them.</p> <h1>Data Preprocessing and Augmentation</h1> <p>Basically the same as the previous 1st solution. But I found using 3d positions (means including depth) and pose landmarks gives better CV and LB score.<br> My input: left-right hand, eye, nose, lips and pose landmarks.</p> <h1>Model</h1> <p>I believe this task would be quite similar to Automatic Speech Recognition (ASR), so I used Conformer <a href="https://arxiv.org/abs/2005.08100" target="_blank">https://arxiv.org/abs/2005.08100</a></p> <h1>Post-preprocessing</h1> <ol> <li>I checked my worst predictions in the validation set and found that shorter predictions are worse. And most very short predictions (length less than 5) are basically predicting nothing, e.g. single characters like "a" or space " ".</li> <li>I found that only few label's length is less or equal to 5. </li> <li>So adding some make-up phrases to very short predictions might be a good idea.</li> <li>I picked the most common chars in training set: "a", "e", "r", "o", "-", " ".</li> <li>I test all the combinations of "a", "e", "r", "o", "-", " " based on validation set and " -aero" is the best.</li> </ol> <h1>What not worked</h1> <ol> <li>Autoregressive transformer models. <br> I spent most of my time on transformers. But they strongly overfits the phrase and I couldn't find a way to solve it.<br> The problem is , for an input X[1…n] and its phrase "123456789". We may expect our model predict "12345" for the first half input X[1…n/2]. <br> For CTC models, it is true. But my autoregressive model always makes half correct prediction "12345" + half random incorrect prediction e.g. some random 5-digit numbers.</li> <li>Masking for variable-length input<br> I tried to use the same masking techniques as the previous 1st solution <a href="https://www.kaggle.com/code/hoyso48/1st-place-solution-training" target="_blank">https://www.kaggle.com/code/hoyso48/1st-place-solution-training</a>, to support longer input frames.<br> However, masking models always produce lower CV and LB scores (-0.02). I believe it is due to the masking limits the output length, but shorter inputs still require a longer output space. Although, unfortunately, I don't have time to verify it, I believe it might enable much better solutions.</li> <li>Empty embedding for fixed-length input<br> I tried to use learnable constant weights for padded empty frames and 2-layer MLP for encoding landmarks in original frames. It seems to me makes more sense, but CV score is worse than direct encoding all frames. </li> </ol> <p>But masking is faster when processing shorter inputs while being able to take longer inputs without resizing. Am I right?</p>
Google - American Sign Language Fingerspelling Recognition	19th Place Solution - Jasper models	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>My solution is quite simple. I wasn't going to write about it, but there are no Jasper-like solutions in discussions yet :)</p> <h1>Preprocessing</h1> <p>Preprocessing was almost the same with <a href="https://www.kaggle.com/competitions/asl-signs/discussion/406306" target="_blank">previous competition</a>. But deleted mixup and replace augmentation. And add more complex time augmentations.</p> <h1>Loss</h1> <p>I only used CTC loss.</p> <h1>Data split</h1> <p>Random split <strong>not</strong> by user for 20 folds grouping by data length.<br> Deleted obviously bad samples:</p> <pre><code>supplemental_metadata_df = supplemental_metadata_df[~((supplemental_metadata_df[] < ) & (supplemental_metadata_df[] > ))] train_df = train_df[~((train_df[] < ) & (train_df[] > ))] </code></pre> <p><br> Used all supplemental data in train.</p> <p>Also tried to create second loader with pseudo labels on bad predicted samples. Or delete bad predicted samples. Used both loaders while training:</p> <pre><code>loader_to_use = epoch % == : loader_to_use = </code></pre> <p>But it didn't help much.</p> <h1>Model</h1> <p>First I tried simple quartznet, but with all strides set to 1.<br> Then I added some features step by step and trained each experiment from previous best checkpoint.</p> <h3>Same for all runs:</h3> <ul> <li>Optimizer: LookAheadAdamW</li> <li>Scheduler: Onecycle</li> <li>Batch size: 80</li> </ul> <h3>List of experiments:</h3> <ul> <li>Simple 5x5 quartznet with stride = 1. [lr = 8e-3] [epochs=200] CV:0.8035</li> <li>Simple 5x5 quartznet with stride = 1. [lr = 7e-3] [epochs=150] CV: 0.8106</li> <li>Previous + SE blocks. [lr = 6e-3] [epochs=171] CV: 0.8160</li> <li>Previous + Deep supervision (DSV) outputs for train loss. [lr = 5.8e-3] [epochs=175] CV: 0.8211</li> <li>Previous without DSV + higher aug probabilities. Note: score worse but it needed to train one checkpoint without DSV for better next train results. [lr = 6e-3] [epochs=180] CV: 0.8177</li> <li>Previous + DSV + increased dropout. [lr = 6.1e-3] [epochs=200] CV: 0.8237</li> <li>Previous - DSV + lower dropout. [lr = 6e-3] [epochs=173] CV: 0.8188</li> <li>Previous + Hyper Column (HYP) + 2 more blocks with 5 repeats. [lr=5.8e-3] [epochs=180] CV: 0.8252</li> <li>Previous + masked 1conv + masked SE + 2D DepthWise CNN layer before classifier + more dropout. [lr=5.8e-3] [epochs=200] CV: 0.8294</li> <li>Previous + lstm after 2D Conv. [lr=1.75e-3] [epochs=125] CV: 0.8301</li> </ul> <h1>What didn't work for me:</h1> <ul> <li>Conformer models. I tried it but I guess without enough effort.</li> <li>Adding attention or conformer blocks to Jasper models.</li> </ul> <p>When I used raw conformer and it has very low score. But when I changed input 2D conv to 1D conv or when I just deleted it - it was much better, but still little worse than my Jasper models.</p> <h1>What I should have done, but didn't:</h1> <ul> <li>Train more epochs</li> <li>Try better with conformer or squeezeformer</li> <li>Add post processing</li> <li>Try seq2seq</li> <li>Fixed length of input</li> </ul> <p>Thanks to Kaggle for this great competition. Thanks to all top participants who shared their awesome solutions. <br> Congratulations to all winners! </p> <blockquote> <p>How to enforce fixed length of input?</p> </blockquote> <p><a href="https://pytorch.org/docs/stable/generated/torch.nn.functional.interpolate.html" target="_blank">TORCH.NN.FUNCTIONAL.INTERPOLATE</a></p> <p>Yes, I developed pytorch, and converted it with nobuco. But didn't use Nemo, wrote my own copying some blocks and logic from Nemo.<br> One of the reasons to write my own - is ability to convert this models with nobuco without a lot of changes</p>
Google - American Sign Language Fingerspelling Recognition	20th place solution	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thanks to Kaggle, Google and other organizers for hosting this exciting competition. Hope this kinds of competition open in kaggle more often.</p> <p><strong>TLDR</strong></p> <p>My solution is a single large model trained with CTC loss with no ensemble(39MB). The output shape of the model is (batch, sentence length, class).</p> <p><strong>Data Preprocessing and Augmentation</strong></p> <p>Basically the same as the previous 1st solution. But I found using (x, y, z) and pose landmarks gives better score. I used the following landmarks: left hand, right hand, eye, lips and pose landmarks.</p> <p><strong>Model</strong></p> <p>I used previous 1st solution. However, I used the branchformer. Also, I changed the multi head attention to relative multi head attention. My model consists of 1 dense layer - 4 conv1d layer - 6 branchformer layer. In addition, I used stochastic depth.</p> <p><strong>Training</strong></p> <p>Epoch = 150<br> bs = 64<br> Lr = 8e-4<br> AWP = Epoch * 0.1<br> Schedule = CosineDecay with warmup ratio 0.1<br> Optimizer = Adam with Lookahead<br> Loss = Inter CTC loss(<a href="https://arxiv.org/abs/2102.03216" target="_blank">https://arxiv.org/abs/2102.03216</a>)</p>
Google - American Sign Language Fingerspelling Recognition	22nd place solution: CTC Loss, Strong augmentations, CNN+MHSA	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>The augmentations are applied randomly during training</p> <p><a href="https://github.com/SamratThapa120/sign-language-finger-spelling/blob/master/signet/models/feature_extractor_downsampled.py" target="_blank">My best-performing model</a> (<a href="https://github.com/SamratThapa120/sign-language-finger-spelling/blob/master/signet/configs/ctc_loss_with_downsampled_deploy_concataug.py" target="_blank">config</a>)is based on the <a href="https://www.kaggle.com/code/hoyso48/1st-place-solution-training/notebook" target="_blank">1st place solution of the previous competition</a> by <a href="https://www.kaggle.com/hoyso48" target="_blank">@hoyso48</a>. I trained the model using vanilla CTC Loss, and used greedy decoding for inference. </p> <h3>Things that worked for me:</h3> <ul> <li><p><strong>Longer input frames length:</strong> A longer input frame length of 384 performed better than shorter length of 256. </p></li> <li><p><strong>Deeper/larger model:</strong> My best-performing model has 7 blocks with hidden dimension of 256. It has 9.5M parameters.</p></li> <li><p><strong>Pose keypoints:</strong> Adding pose information was helpful. Best-performing model uses pose+hands+lips+eyes.</p></li> <li><p><strong>CNN+MHSA:</strong> CNN+MHSA model > Only CNN Model>Only MHSA model</p></li> <li><p><strong>Strong augmentation:</strong> My model was performing well based on local CV, and public LB, with strong correlations between the two.But when I tested the model using this <a href="https://github.com/SamratThapa120/gradio-ASL-fingerspelling-recognition" target="_blank">Gradio app</a> using my webcam, it could not recognize my signs, so I had to use strong data augmentations to get the model to work. Especially temporal mask helped because I sign slower than the pros. These augmentations also boosted the public LB score by +0.006.</p> <p>flip_lr_probability=0.5<br> random_affine_probability=0.75<br> freeze_probability=0.5<br> temporal_mask_probability=0.75<br> temporal_mask_range=(0.2,0.4)</p></li> <li><p><strong>concat augmentation:</strong> Randomly concatenate two short landmark sequences, as well as their labels. This improved public LB by +0.008. I applied this augmentation to 40% of all training samples.</p></li> </ul> <h3>Here are other things I tried, that did not contribute to the best-performing model:</h3> <ul> <li><p><strong>transformer-encoder+decoder:</strong> Transformer endoder-decoder model with cross-entropy loss.</p></li> <li><p><strong>transformer-decoder:</strong> CNN+MHSA model and Transformer-like decoder with cross-entropy loss.</p></li> <li><p><strong>Causal-masking in self-attention:</strong> removing causal masking performed better than using causal mask.</p></li> <li><p><strong>Attention-span in self-attention:</strong> I thought that there would not be long-term dependency between frames for this task, so I tried to reduce the attention span of self-attention. Although there was no performance degradation, it did not boost performance either.</p></li> <li><p><strong>Focal loss</strong>: I tried the CTC Focal loss based on <a href="https://github.com/TeaPoly/CTC-OptimizedLoss/blob/main/ctc_focal_loss.py" target="_blank">this</a> repo, but there were no gains.</p></li> <li><p><strong>Erase landmarks augmentation</strong>: Randomly erase landmarks except the hand landmarks, to make the model more robust to mediapipe's detection errors.</p></li> </ul> <p>Originally I was using <a href="https://github.com/SamratThapa120/sign-language-finger-spelling/tree/pytorch" target="_blank">pytorch</a>, but I switched to Tensorflow because I faced several issues when converting the pyTorch model to tfLite. However, training with CTC Loss was upto 10x slower in tensorflow than pytorch. Looking back, I think I should have put more effort into fixing the model conversion issue, as I would have been able to perform more experiments. </p> <p>Congratulations to the winning teams. I would also like to thank Google for hosting this competition, it was a valuable opporunity to learn many new things. Also, I would like to thank everyone who shared their notebooks and unique ideas.</p> <p>Freeze augmentation tries to replicate frozen frames in a video. Randomly select some frames and replicate them without changing their order in the sequence. For example [a,b,c,d,e] will get converted to something like [a,a,a,b,b,c,d,e,e]. </p> <p><a href="https://github.com/SamratThapa120/sign-language-finger-spelling/blob/2bbc9a4c51622d3553ab9be17715ca51332c8797/signet/dataset/transforms.py#L215" target="_blank">Here is the code</a></p>
Google - American Sign Language Fingerspelling Recognition	25th Place Solution (time warpping + relative positional encoding + intermediate CTC)	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thanks to Kaggle and Google for hosting this competition. This is my first Kaggle competition experence and I have learnt a lot from the previous ASL competition and other participants. I start my model from <a href="https://www.kaggle.com/hoyso48" target="_blank">@hoyso48</a>'s solution from the last competition and added some components:</p> <p><strong>Data Augmentation</strong><br> Applied random time warpping (LB +0.01)</p> <pre><code>def temporal: n = tf.shape(x) n < : return x w = n t1 = tf.clip, , n-) t2 = tf.clip + t1, , n-) new_x1 = interp1d new_x2 = interp1d return tf.concat(, axis=) def augment: tf.random.uniform()< always: x = resample(x, (,)) tf.random.uniform()< always: x = flip max_len is not None: x = temporal tf.random.uniform()< always: x = spatial tf.random.uniform()< always: x = temporal tf.random.uniform()< always: x = temporal tf.random.uniform()< always: x = spatial return x </code></pre> <p><strong>Data Preprocessing</strong><br> Truncate <code>phrase</code> to have smaller length than the input</p> <pre><code> inp_len <= label_len: offset = tf((), , label_len-inp_len+, dtype=tf.int32) = label_len = inp_len- </code></pre> <p><strong>Multi-headed attention with relative positional encoding</strong> (LB +0.02)</p> <pre><code> : def : super. self.dim = dim self.head_dim = dim self.scale = self.head_dim- self.num_heads = num_heads self.num_nbr = num_nbr self.qkv = tf.keras.layers. self.drop1 = tf.keras.layers. self.proj = tf.keras.layers. self.supports_masking = True self.wgt_v = self.add, ='glorot_uniform',trainable=True, name=+str(tf.keras.backend.get)) self.wgt_k = self.add, ='glorot_uniform',trainable=True, name=+str(tf.keras.backend.get)) def : idx_mat = tf.reshape(tf.tile(tf.range(length), ), ) idx_mat = idx_mat - tf.transpose(idx_mat) return tf.clip + self.num_nbr def : xy = tf.matmul(x, y, transpose_b=transpose) x = tf.keras.layers.)(x) mul = tf.matmul(x, z, transpose_b=transpose) mul = tf.keras.layers.)(mul) return xy + mul def call(self, inputs, mask=None): mask is not None: mask = mask length = inputs.shape idx_mat = self. rpe_k = tf.gather(self.wgt_k, idx_mat) rpe_v = tf.gather(self.wgt_v, idx_mat) qkv = self.qkv(inputs) qkv = tf.keras.layers.)(tf.keras.layers.)(qkv)) q, k, v = tf.split(qkv, , axis=-) logits = self.self.scale attn = tf.keras.layers.(logits, mask=mask) attn = self.drop1(attn) x = self. x = tf.keras.layers.)(tf.keras.layers.)(x)) x = self.proj(x) return x </code></pre> <p><strong>Intermediate CTC</strong></p> <pre><code>def get_base_model(=384, =384): inp = tf.keras.Input((max_len,CHANNELS), =) x = tf.keras.layers.Masking(=PAD,input_shape=(max_len,CHANNELS))(inp) x = tf.keras.layers.Dense(dim, =,name='stem_conv')(x) x = tf.keras.layers.BatchNormalization(=0.95,name='stem_bn')(x) ksize = 17 drop_rate = 0.2 x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) inter = tf.keras.layers.Dense(dim, =None)(x) inter = tf.keras.layers.Dense(NUM_CLASSES+1)(inter) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = Conv1DBlock(dim,ksize,=drop_rate)(x) x = TransformerBlock(dim,=2)(x) x = tf.keras.layers.Dense(dim,=None,name='top_conv')(x) x = tf.keras.layers.Dense(NUM_CLASSES+1, =)(x) return tf.keras.Model(inp, [inter, x]) def ctc_loss(args): y_pred, inp_len, label, label_len = args return tf.nn.ctc_loss(label, y_pred, label_len, inp_len, =, =NUM_CLASSES) def get_model(base_model): label = tf.keras.Input(shape=(MAX_LEN_OUTPUT,), =, =) label_len = tf.keras.Input(shape=(), =, =) inp_len = tf.keras.Input(shape=(), =, =) loss1 = tf.keras.layers.Lambda(ctc_loss, output_shape=())( [base_model.output[0], inp_len, label, label_len]) loss2 = tf.keras.layers.Lambda(ctc_loss, output_shape=())( [base_model.output[1], inp_len, label, label_len]) loss = 0.3 * loss1 + 0.7 * loss2 #loss = 0.0 * loss1 + 1.0 * loss2 model = tf.keras.Model( inputs=[base_model.input, inp_len, label, label_len], =loss) return model </code></pre> <p><strong>Fine tuning</strong> (LB +0.005)<br> The model was first trained with both training data and supplementary data with intermediate CTC (0.3 and 0.7) for 300 epochs. Then the pre-trained model was fine tuned only on the training data without intermediate CTC (0.0 and 1.0)</p> <p>I will post the training notebook soon.</p> <p><a href="https://www.kaggle.com/hoyso48" target="_blank">@hoyso48</a>'s solution<br> <a href="https://www.kaggle.com/code/hoyso48/1st-place-solution-training" target="_blank">https://www.kaggle.com/code/hoyso48/1st-place-solution-training</a></p>
Google - American Sign Language Fingerspelling Recognition	29th Place Solution - LB 0.762	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h2>TL;DR</h2> <p>We used a 1D convolutional network with 122-dimensional features as input and Self Conditional CTC, along with a greedy search as the decoder.</p> <h2>Preprocess</h2> <p>We used xy feature points of the hand performing the gesture and xy feature points of the lips, for a total of 61 xy feature points (= 122-dimensional vector).</p> <h2>Data Augmentation</h2> <p>Overfitting was suppressed by randomly dropping hand feature points. The percentage of dropping was 40% of the total, and at least 5 consecutive frames were dropped.</p> <h2>Model</h2> <p>We used a 1DCNN consisting of 9 Residual Blocks of kernel width 3 and 12 Residual Blocks of kernel width 31.</p> <h2>Loss Function</h2> <p>We used Self Conditional CTC, a variant of CTC, for the loss function. By using Self Conditional CTC, a significant performance improvement was seen without increasing the amount of calculation.<br> <a href="https://arxiv.org/abs/2104.02724" target="_blank">Read more about Self Conditional CTC here</a></p> <h2>Training</h2> <ul> <li>60 epochs</li> <li>CosDecay learning rate</li> <li>AdamW optimizer</li> <li>AWP</li> </ul> <h2>Convert Model</h2> <p>All implementation was done in PyTorch and converted to 16-bit quantized tflite by onnx2tf. Conversion using nobuco was also tried, but was not adopted because, contrary to expectations, it increased processing time.</p> <h2>Not Worked</h2> <ul> <li>Stochastic Weight Averaging was very effective in the last competition but did not contribute to performance in this competition.</li> <li>We thought that Levenshtein OCR could be used as a decoder and tried to implement it, but it did not work.</li> </ul>
Google - American Sign Language Fingerspelling Recognition	2nd place solution - Test & Compare ASR algorithms	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>I used KaggleDatasets(same as previous competition, converted .parquet to .tfrecords). every kaggle dataset has its own Google Cloud Storage path, so if it is public and you have the actual gcs path then you can access it anywhere with GCS.</p> <p>Thanks and congratulations! Your work through these two competitions was very masterful. need more time to digest, but I've already learned a lot from your solution :)</p> <p>Once again, thanks to Kaggle, Google and other organizers for hosting this exciting competition. I felt this one as well as the previous competition, is well-organized so that we can try various ideas and could learn a lot from each other. Hope this kinds of competition open in kaggle more often.. :)</p> <h2>TLDR</h2> <p>My overall solution is almost identical to the previous competition. It's mainly just the adoption of ASR(Automatic Speech Recognition) algorithms (joint CTC + Attention) on top of the previous 1st place model. The details of the model, preprocessing, and training are largely similar to the previous competition, so please refer to the <a href="https://www.kaggle.com/competitions/asl-signs/discussion/406684" target="_blank">previous competition solution</a> for more details on model/training/augmentations/etc.</p> <h2>ASR Algorithms</h2> <p>I had a deep interest in ASR, and after recognizing its correlation with the competition, I studied and experimented with various algorithms. While I couldn't find anything particularly superior to the widely used vanilla CTC among the numerous ASR algorithms, I learned a lot through various experiments, and I will mainly share those insights.</p> <p>As I began studying ASR for this competition, I discovered from reading papers that current NN ASR algorithm mainly consists of CTC, Attention-based, and Transducer. I implemented all three (though there are more diverse algorithms out there). In conclusion, from a baseline performance perspective, all three algorithms were quite similar and each have pros & cons. Here are the insights I gathered from implementing each algorithm:</p> <ul> <li><p>CTC:</p> <ul> <li>During greedy decoding, the prediction step = O(1). Therefore, when using a Single model, GreedyCTC is the most efficient. (this is probably why many solutions use a large single model with CTC).</li> <li>With beam search, prediction step = O(L) (where L = encoder output length). Depending on the efficiency of the algorithm, the computational overhead isn't very large, so the overhead by introducing beam search isn't very significant. However, the performance improvement from using beam search is minimal (+0.003), making it not very tempting(as its tflite-convertible implementation is not quite trivial) compared to Attention.</li> <li>As discussed, there's no guarantee of alignment between model prediction timesteps, so simple average ensembling can't be applied.</li></ul></li> <li><p>Attention-based:</p> <ul> <li>Uses an autoregressive approach, so even with greedy decoding, prediction step = O(N) (where N = decoder output length).</li> <li>If the decoder is an RNN or Transformer, stateful inference(i.e. previous key, value caching with Transformer) can be used to reduce the complexity. In actual implementation with Transformer, it reduced the inference time on the CPU by about 20~30%.</li> <li>Beam search is possible, but unlike CTC, it requires introducing appropriate heuristics to penalize the output length. Although several attempts were made, none worked. </li> <li>It's easier to apply ensembling with Attention. Simply take the average of each model at each prediction step, and it works well (ensembling three models gave +0.009).</li> <li>More room for score improvement than CTC(ex more decoder layers with augmentations).</li></ul></li> <li><p>Transducer:<br> Prediction steps = O(L) (where L = encoder output length). Due to the most prediction steps in a greedy manner, it's not very efficient. I didn't consider it further as optimization was harder and performance was slightly lower compared to CTC or Attention. However, it has the advantage of real-time recognition in streaming mode if the encoder is causal (which wasn't relevant for this competition).</p></li> </ul> <p>Rough single model inference time on test dataset with kaggle kernel is as follows.</p> <blockquote> <p>CTC greedy(~40mins) < AttentionGreedy=CTCBeamsearch(~1h20mins) <= CTCAttentionJointGreedy(~1h30mins)</p> </blockquote> <h2>CTC-Attention Joint Training & Decoding</h2> <p>As both CTC and Attention showed similar performance, I tried to find a method to utilize both techniques. I primarily referred to the following two papers:</p> <p><em>Joint CTC-Attention Based End-To-End Speech Recognition Using Multi-Task Learning, Kim et al. 2017.</em><br> <a href="url" target="_blank">https://arxiv.org/pdf/1609.06773.pdf</a><br> <em>Joint CTC/attention decoding for end-to-end speech recognition, Hori et al. 2017.</em><br> <a href="url" target="_blank">https://aclanthology.org/P17-1048.pdf</a></p> <p>Joint CTC-Attention Training is, as the name implies, adding both a CTC decoder (single GRU layer) and an attention decoder (single Transformer decoder layer) to one encoder for multitask learning. The loss weight is set to CTC=0.25 and attention=0.75. But the Joint training itself did not bring noticeable performance boost.</p> <p>By using the CTC prefix score, it is possible to calculate the probability of an arbitrary output hypothesis "h" without depending on the output timestep of the CTC output. In other words, <strong>implementing the CTC prefix score computation allows for ensembling outputs not only between CTC models but also between CTC and Attention models</strong>. CTC-attention joint decoding with CTC weight=0.3 showed a performance improvement of +0.007~8 without much impact on inference time. It was especially challenging for me to implement it accurately, efficiently, and without any issues to be compatible with tflite.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5003978%2Fabde23b44b90f8ba6adc7d56ca874d86%2F.drawio-3.svg?generation=1692977029060458&alt=media" alt=""></p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5003978%2F22c798e256f66322a0cc9e42b976d570%2F.drawio-4.svg?generation=1692977118893507&alt=media" alt=""></p> <h2>Preprocessing</h2> <p>Similar to the previous competition solution, but simpler. Every landmark and xyz was used and standardized. Flip left-handed signer(rather than augmentation). MAX_LEN=768 was used. Any hand-crafted feature was not significant, likely due to the absence of complex relations between movements in frames.</p> <h2>Model</h2> <p>Encoder is the same as the previous competition (Stacked Conv1DBlock + TransformerBlock) but increased in size (expand ratio 2->4 in Conv1DBlock) and depth (8 layers -> 17 layers). Single model has ~6.5M parameters. I applied padding='same'(rather than 'causal') and output stride=2 (requires slightly more logic for handling masking). In my case, mixing in Transformer blocks wasn't as effective as in the previous competition. Perhaps global features were less crucial in this comp. Additionally, I added one BN to input of the Conv1DBlock for more training stability(especially with awp + more epochs).</p> <p>CTC decoder used a single GRU layer followed by one FC layer. Attention Decoder used a single-layer Transformer decoder. Introducing augmentation to the Decoder input and adding up to 4 Decoder layers improves the performance of the attention decoder (up to +0.004). However, considering the number of parameters and inference speed, I considered it inefficient and thus used a single-layer decoder.</p> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5003978%2F7d1f7aee69242af5e6aa5ef19c9460f6%2Fmodeldesign.drawio-2.svg?generation=1692993585923781&alt=media" alt=""></p> <h2>Augmentations</h2> <ul> <li>Random resample (0.5x ~ 1.5x to original length)</li> <li>Random Affine</li> <li>Random Cutout</li> <li>Random token replacement on decoder input(prob=0.2)</li> </ul> <p>What I overlooked this time is that augmentations which showed no performance improvement or even degraded performance in shorter epochs might actually help improve performance in longer epochs. I experienced a similar phenomenon in the previous competition, but it seemed more pronounced in this one. By selecting augmentations solely based on the results from a 60epoch experiment, I think I missed out on many potentially beneficial augmentations when testing the 400epoch training in the final week.</p> <h2>Training</h2> <p>Epoch = 400<br> bs = 16 * num_replicas = 128<br> Lr = 5e-4 * num_replicas = 4e-3<br> AWP = 0.2 starts at 0.1 * Epoch<br> Schedule = CosineDecay with warmup ratio 0.1<br> Optimizer = AdamW (slightly better than RAdam with Lookahead)<br> Loss = CTC(weight=0.25) + CCE with label smoothing=0.1~0.25(weight=0.75)</p> <p>Training takes around 14 hours with colab TPUv2-8(as colab TPU runtime recently reduced to 3~4 hours, needed 4 consecutive sessions to complete training).<br> Longer Epoch always gave better CV(5fold split by id) and LB but got no time to try over 400 epochs.</p> <h2>LB history</h2> <table> <thead> <tr> <th></th> <th>public LB</th> <th>private LB</th> </tr> </thead> <tbody> <tr> <td>prevcompsinglemodel + CTC (or Attention)</td> <td>0.76</td> <td>0.74</td> </tr> <tr> <td>+ deeper and wider model, add pose</td> <td>0.79</td> <td>0.78</td> </tr> <tr> <td>+ 3 seed ensemble with Attention</td> <td>0.80</td> <td>0.79</td> </tr> <tr> <td>+ ctc attention joint decoding</td> <td>0.81</td> <td>0.80</td> </tr> <tr> <td>+ use all landmarks, longer epoch</td> <td>0.82</td> <td>0.81</td> </tr> </tbody> </table> <p>Seeing solutions from other kagglers, not only the top solutions but also the public notebooks is always inspiring. Thanks to kagglers who shared their insights and ideas as always. And big congrats to <a href="https://www.kaggle.com/christofhenkel" target="_blank">@christofhenkel</a> and <a href="https://www.kaggle.com/darraghdog" target="_blank">@darraghdog</a> for winning this highly competitive competition!</p> <p>Training/Inference code: <a href="https://www.kaggle.com/competitions/asl-fingerspelling/discussion/436873" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/discussion/436873</a></p> <p>Thanks, and congrats on winning! I've always learned from top Kagglers like you, and now I'm happy that I could also provide you with some insights :)</p> <p>Glad to know that my prev solution gave you some help :)<br> I relied on my prev model and gave minimal changes simply because it's my own implementation from prev competition with lots of trials&errors..haha<br> Definitely should study&try more recent conv + transformer hybrid architectures next time. Thanks!</p> <p>Yes. I only used Colab TPUv2-8.</p> <p>I did not refer to any particular paper in the previous competition. Of course, I did search through some papers on related tasks (such as action recognition and GCN models) at the beginning, but I realized that the methods from those papers were not very effective.</p>
Google - American Sign Language Fingerspelling Recognition	33rd place solution - LB 0.758	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thank you :)<br> Yes, we implemented it as you mentioned, but it didn't work.<br> There are moving gesture characters such as 'z' and we guess reversed motion may interrupt learning.<br> Augmentation of phrase consiting with only static gesture may helpful but we didn't try it</p> <h2>TL;DR</h2> <p>We used a conformer-like model consisting transformer encoder, 1d convolution with CBAM and bi-LSTM.<br> Overall model size is about 16MB after INT8 quantization.<br> Training objective is CTC with InterCTC loss.</p> <h2>Data preprocess</h2> <ul> <li>Used landmarks<ul> <li>1 for nose, 21 for dominant hand, 40 for lips</li> <li>x, y coordinates</li></ul></li> <li>Normalization<ul> <li>Standardized distance from nose coordinates</li></ul></li> <li>Feature engineering<ul> <li>Concatenation of normalized location, difference of next frame, joint distance of hand</li> <li>Total 582 dims</li></ul></li> <li>Removing data that input frame is shorter than 2 times of target phrase</li> </ul> <h2>Data augmentation</h2> <ul> <li>Horizontal flip landmarks</li> <li>Interpolation</li> <li>Affine transfrom</li> </ul> <h2>Model</h2> <ul> <li>2 stacked encoder with Transformer, 1D convolution with CBAM and Bi-LSTM<ul> <li>hidden dim: 352</li></ul></li> <li>CTC loss and Inter CTC loss after first encoder</li> <li>17M parameters and INT8 quantization</li> </ul> <h2>Train</h2> <ul> <li>2 staged training (300 + 200 epochs)<ul> <li>Use supplemental and train data in first 300 epochs</li> <li>Use train data 200 epochs</li></ul></li> <li>Ranger optimizer</li> <li>Cosine decay scheduler with 12 epochs warmup</li> <li>AWP<ul> <li>It prevents the validation loss from diverging, but it doesn't seem to improve the edit distance.</li> <li>Used to train long epochs.</li></ul></li> </ul> <h2>Not worked</h2> <ul> <li>Augmentation<ul> <li>Time, spatial, landmark masking</li> <li>Time reverse</li></ul></li> <li>Autoregressive decoder<ul> <li>Use joint loss, CTC loss for encoder and Crossentropy for decoder</li> <li>Inference time is longer and the number of parameters larger than those of the CTC encoder alone, but the performance improvement is not clear, so it is not used.</li></ul></li> </ul>
Google - American Sign Language Fingerspelling Recognition	449th Position Solution for Google ASLFR Competition - Adversarial Regularization, Quantization Aware Training, and KD Code for Tensorflow!	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Hi Everyone! I'm just posting my solution here, in case it could help someone! I used a lot of techniques, but didn't have the computing power to push the boundaries required for a very high rank. Nevertheless, I learnt a lot of techniques, and am writing this solution here to help provide useful code!</p> <h1>Competition Context</h1> <p>The context of this solution is the Google - American Sign Language Fingerspelling Recognition Competition:<br> <a href="https://www.kaggle.com/competitions/asl-fingerspelling/overview" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/overview</a></p> <p>The data of this competition can be found here:<br> <a href="https://www.kaggle.com/competitions/asl-fingerspelling/data" target="_blank">https://www.kaggle.com/competitions/asl-fingerspelling/data</a></p> <h1>My Approach</h1> <p>My approach was identical to most people, I referred to the really awesome notebook by Rohit Ingilela here: <a href="https://www.kaggle.com/code/irohith/aslfr-ctc-based-on-prev-comp-1st-place" target="_blank">https://www.kaggle.com/code/irohith/aslfr-ctc-based-on-prev-comp-1st-place</a></p> <p>I made several additions and changes, namely by adding Quantization Aware Training, AWP, and Knowledge distillation (not necessarily in the same model). I tested a lot of parameters as well and made a lot of models.</p> <p>My idea was to use Quantization to increase model size. I found that after 24 Million parameters, even though the model size was small enough, the performance became too poor to run within the 5 hours allocated for testing. I found that KD didn't improve my performance too much, so I didn't explore it a lot. I tried AWP only at the end, and even though it improved performance by around 0.003, I didn't get to test it a lot.</p> <p>Nevertheless, here is some code for each idea:</p> <p><strong>Quantization:</strong></p> <pre><code>!pip install -q tensorflow-model-optimization tensorflow_model_optimization tfmot tensorflow.keras layers x = tfmot.quantization.keras.quantize_annotate_layer(layers.Dense(channels, activation = activation, kernel_regularizer = regularizer))(inputs) tflitemodel_base = TFLiteModel(model) tflitemodel_base(frames)[].shape keras_model_converter = tf.lite.TFLiteConverter.from_keras_model(tflitemodel_base) keras_model_converter.optimizations = [tf.lite.Optimize.DEFAULT] keras_model_converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS] tflite_model = keras_model_converter.convert() (, ) f: f.write(tflite_model) </code></pre> <p><strong>Adversarial Regularization in Training</strong></p> <pre><code>! pip install -q neural-structured-learning neural_structured_learning nsl adv_config = nsl.configs.make_adv_reg_config( multiplier = , adv_step_size = , adv_grad_norm = ) (): {: image, : label} train_dataset = train_dataset.(convert_to_dictionaries) val_dataset = val_dataset.(convert_to_dictionaries) strategy.scope(): model = nsl.keras.AdversarialRegularization( model, label_keys = [], adv_config = adv_config ) loss = CTCLoss optimizer = tfa.optimizers.RectifiedAdam(sma_threshold = ) optimizer = tfa.optimizers.Lookahead(optimizer, sync_period = ) model.(loss = loss, optimizer = optimizer) </code></pre> <p><strong>Knowledge Distillation</strong></p> <pre><code> (keras.Model): (): ().__init__() self.student = student self.teacher = teacher (): ().(optimizer=optimizer) self.loss = CTCLoss self.distiller_loss_fn = keras.losses.KLDivergence() self.alpha = alpha self.temp = temperature (): x, y = data is_batch_small = tf.math.less(tf.shape(x)[], train_batch_size) is_batch_small = tf.math.reduce_all(is_batch_small) (): tf.constant(), tf.constant() (): teacher_pred = self.teacher(x, training = ) tf.GradientTape() tape: student_pred = self.student(x, training = ) student_loss = self.loss(y, student_pred) distilled_loss = self.distiller_loss_fn( tf.nn.softmax(student_pred / self.temp), tf.nn.softmax(teacher_pred / self.temp) ) * (self.temp ** ) loss = self.alpha * student_loss + ( - self.alpha) * distilled_loss trainable_vars = self.student.trainable_variables gradients = tape.gradient(loss, trainable_vars) self.optimizer.apply_gradients((gradients, trainable_vars)) self.compiled_metrics.update_state(y, student_pred) student_loss, distilled_loss student_loss, distilled_loss = tf.cond(is_batch_small, pass_batch, normal) results = {m.name : m.result() m self.metrics} results.update( {: student_loss, : distilled_loss} ) results (): () x, y = data is_batch_small = tf.less(tf.shape(x)[], val_batch_size) is_batch_small = tf.reduce_all(is_batch_small) (): tf.constant(), tf.constant() (): y_pred = self.student(x, training = ) teacher_pred = self.teacher(x, training = ) student_loss = self.loss(y, y_pred) distilled_loss = self.distiller_loss_fn( tf.nn.softmax(y_pred / self.temp), tf.nn.softmax(teacher_pred / self.temp) ) * (self.temp ** ) self.compiled_metrics.update_state(y, y_pred) student_loss, distilled_loss student_loss, distilled_loss = tf.cond(is_batch_small, pass_batch, normal) results = {m.name : m.result() m self.metrics} results.update( {: student_loss, : distilled_loss} ) results </code></pre> <p>While I didn't get a super high rank, I did learn a lot and used a plethora of techniques! I wrote most things from scratch, to learn things myself, and referred to other's work only to get a better understanding of what they did and improve my own code as a result.</p> <p><em>Validation</em><br> Quite standard - kept a holdout set and computer Normalized Levenshtein distance on it after every epoch.<br> No CV due to low computational resources.</p> <h1>Improvements</h1> <p>I am quite disappointed that the things I missed out were very trivial - such as not training for more than 300 epochs (I could only manage 100 to try enough models and variations out) and that keeping some missing hands frames was important.</p> <p>I also didn't use enough augmentations which was important, even though I used all the ones used by top submissions except cutout.</p> <p>I also read about Squeezeformer, but got demotivated because I saw high scoring public submissions without Squeezeformer :(</p> <p>To further improve, I'll correct all the points above.</p> <h1>Final Note</h1> <p>I am quite happy to have participated in this fun and awesome contest. I really learnt a lot, and experimented a lot. I am certain that what I did here will help me in future competitions, research and work.</p> <p>If you could give me any suggestions/feedback on my code above, please do so. It will help me improve and learn, and I will be very grateful for your help!</p> <p>Thank you so much!</p>
Google - American Sign Language Fingerspelling Recognition	95th Bronze solutions	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><h1>95th Bronze solutions</h1> <p>Thanks to Kaggle for hosting this interesting competition!!!!<br> This competition was diffucult for us…<br> There were many differences from last competition which made it challenging. I enjoyed learning about transcription task.</p> <h1>Summary</h1> <p>We utilized the best public notebook with some modifications.</p> <ul> <li>base notebook: <a href="https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-place" target="_blank">https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-place</a></li> </ul> <h1>What changes were made from the public notebook?</h1> <ul> <li><p>Normalization (+0.002 LB)</p> <ul> <li>Adjusted all coordinates so that the center of numbers 11 and 12 is set to 0.<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2930242%2F6d424a3976cee8f769e3e90ba54a9ac2%2Finbox_2930242_fa394dbd96e33fb875aef6ed1f25c757_1.png?generation=1692922021894426&alt=media" alt=""></li></ul></li> <li><p>Added more features (+0.01 LB)</p> <ul> <li>Motion, Shape Feature, <ul> <li>Shape is distance of this point<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2930242%2F3f7fdc56a208576b7664408fd807e422%2Finbox_2930242_b739316822c577ef9b4be204325f569e_2.png?generation=1692922063719858&alt=media" alt=""></li></ul></li></ul></li> <li><p>Feature Augmentation (+0.01 LB)</p> <ul> <li>Framerate, randomDrop, random padding, random crop.</li></ul></li> <li><p>Changed Model (+0.005 LB)</p> <ul> <li>dim : 384 => 192</li> <li>conv trans Block num : 3=>10</li> <li>Add aux Loss</li></ul></li> <li><p>Pretraining with supplementay data (+0.01 LB)</p></li> <li><p>Training params</p> <ul> <li>epoch 100</li> <li>batchsize 256</li> <li>learningrate 1e-2</li></ul></li> </ul> <h1>Things I couldn't do</h1> <ul> <li>Error correction Network.<ul> <li>I read the following paper and tried to incorporate voice recognition technology, but I couldn't make it in time.</li> <li><a href="https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjAkrKRvvaAAxWBO3AKHbDkB4MQFnoECBMQAQ&url=https%3A%2F%2Farxiv.org%2Fabs%2F2111.01690&usg=AOvVaw3B--e8pJqXTaYolWYXIBOf&opi=89978449" target="_blank">https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjAkrKRvvaAAxWBO3AKHbDkB4MQFnoECBMQAQ&url=https%3A%2F%2Farxiv.org%2Fabs%2F2111.01690&usg=AOvVaw3B--e8pJqXTaYolWYXIBOf&opi=89978449</a></li></ul></li> <li>Change to SOTA OCR model.<ul> <li><a href="https://arxiv.org/pdf/2205.00159.pdf" target="_blank">https://arxiv.org/pdf/2205.00159.pdf</a></li></ul></li> </ul> <h1>Appendix: Model code</h1> <pre><code>def get: inp = tf.keras. x = tf.keras.layers.(inp) x = tf.keras.layers.(x) + positional x = tf.keras.layers.(x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x_aux = (x) x = (x_aux) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) x = (x) # main x = tf.keras.layers.(x) x = tf.keras.layers.(x) x = tf.keras.layers.,name='out')(x) #aux x_aux = tf.keras.layers.(x_aux) x_aux = tf.keras.layers.(x_aux) x_aux = tf.keras.layers.,name='aux_out')(x_aux) model = tf.keras. loss = { 'out':CTCLoss, 'aux_out':CTCLoss } # Adam Optimizer optimizer = tfa.optimizers. optimizer = tfa.optimizers. model.compile(loss=loss, optimizer=optimizer) return model </code></pre>
Google - American Sign Language Fingerspelling Recognition	95th Bronze solutions: Focal loss + KD + Supplemental data	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Thank you, Kumar. Utilizing TPU would result in quicker performance compared to a GPU.</p> <h2><strong>95th Bronze Solutions (public score 0.714, private score 0.671)</strong></h2> <p>I express my gratitude to Kaggle and Google for organizing this competition. Participating in this competition has been an invaluable learning experience for me.</p> <p>One significant achievement was improving the public score from Version 17 of Saidineshpola's notebook, which initially achieved a public score of 0.689. The improvements were achieved by implementing the following modifications:</p> <h1><strong>Model & Config:</strong></h1> <ul> <li>Set model dimensions (Dim) to 192.</li> <li>Set 16 CNN-Transformer blocks.</li> <li>Incorporated a dropout rate of 0.4.</li> <li>Utilized 8 attention heads and an expansion factor of 8 in the Transformer blocks.</li> <li>The total number of parameters amounted to 19,255,980.</li> <li>Employed tf.float16 for model conversion.</li> <li>Configured training with 100 epochs, with 10 warm-up epochs.</li> <li>Set the maximum learning rate (LR_MAX) to 1e-3.</li> <li>Set the weight decay (WD_RATIO) to 1e-4.</li> </ul> <h1><strong>CTC loss:</strong></h1> <p>I utilized the CTC loss function from the <code>pip install tf-seq2seq-losses</code> library, which provided comparable results to TensorFlow's built-in function. This implementation offered a speed enhancement of around 25% on V100 GPUs. Additionally, I applied the Focal loss to the CTC loss function.</p> <pre><code> tf_seq2seq_losses classic_ctc_loss (): label_length = tf.reduce_sum(tf.cast(labels != pad_token_idx, tf.int32), axis=-) logit_length = tf.ones(tf.shape(logits)[], dtype=tf.int32) * tf.shape(logits)[] ctc_loss = classic_ctc_loss( labels=tf.cast(labels, dtype=tf.int32), logits=logits, label_length=label_length, logit_length=logit_length, blank_index=pad_token_idx, ) p= tf.exp(-ctc_loss) alpha = gamma = focal_ctc_loss= tf.multiply(tf.multiply(alpha,tf.((-p),gamma)),ctc_loss) loss = tf.reduce_mean(focal_ctc_loss) loss </code></pre> <h1><strong>Data augmentation:</strong></h1> <p>No data augmentation techniques were applied.</p> <h1><strong>Additional data:</strong></h1> <p>Supplemental data was integrated into the training process. I followed <a href="https://www.kaggle.com/code/irohith/aslfr-preprocess-dataset/notebook" target="_blank">ROHITH INGILELA's notebook</a> to generate supplemental TFrecord files. The resources used are available <a href="https://www.kaggle.com/datasets/kongpasom/aslfr-sup-tfrecord" target="_blank">here</a> and <a href="https://www.kaggle.com/datasets/kongpasom/aslfr-sup" target="_blank">here</a> .</p> <h1><strong>Training Approach:</strong></h1> <ol> <li>Trained the model using supplemental data for 100 epochs.</li> <li>Trained the model with training data and the pre-trained weights derived from the - supplemental data for an additional 100 epochs (resulting in a public score of 0.713).</li> <li>Executed a fine-tuning process using knowledge distillation, refining the model with the best pre-trained weights through the AdamW optimizer, resulting in a modest score improvement of +0.001.</li> </ol> <h1><strong>Model Architecture:</strong></h1> <p>Here's a summarized version of the model architecture:</p> <pre><code> (): inp = tf.keras.Input(INPUT_SHAPE) x = tf.keras.layers.Masking(mask_value=)(inp) x = tf.keras.layers.Dense(dim, use_bias=, name=)(x) pe = tf.cast(positional_encoding(INPUT_SHAPE[], dim), dtype=x.dtype) x = x + pe x = tf.keras.layers.BatchNormalization(name=)(x) num_blocks = drop_rate = i (num_blocks): x = Conv1DBlock(dim, , drop_rate=drop_rate)(x) x = Conv1DBlock(dim, , drop_rate=drop_rate)(x) x = Conv1DBlock(dim, , drop_rate=drop_rate)(x) x = TransformerBlock(dim=dim, num_heads=, expand=, attn_dropout=, drop_rate=)(x) x = tf.keras.layers.Dense(dim,activation=,name=)(x) x = tf.keras.layers.Dropout()(x) x = tf.keras.layers.Dense((char_to_num), name=)(x) model = tf.keras.Model(inp, x) loss = CTCLoss optimizer = tfa.optimizers.RectifiedAdam(sma_threshold=) optimizer = tfa.optimizers.Lookahead(optimizer, sync_period=) model </code></pre> <h1><strong>Knowledge Distillation Training :</strong></h1> <p>A training step using knowledge distillation was employed to refine the model:</p> <pre><code>loss_fn = CTCLoss kld_loss_fn = tf.keras.losses.KLDivergence() (): temperature = alpha = teacher_pred = model_teacher(x, training=) tf.GradientTape() tape: logits = model(x, training=) loss_value = loss_fn(y, logits) kld_loss = kld_loss_fn( tf.nn.softmax(teacher_pred / temperature, axis=), tf.nn.softmax(logits / temperature, axis=), ) temperature** loss = alpha * loss_value + ( - alpha) * kld_loss grads = tape.gradient(loss, model.trainable_weights) model.optimizer.apply_gradients((grads, model.trainable_weights)) loss, loss_value, kld_loss </code></pre> <h1><strong>Reference (</strong>If I have seen further, it is by standing on the shoulders of giants.<strong>) :</strong></h1> <p><a href="https://www.kaggle.com/code/irohith/aslfr-ctc-based-on-prev-comp-1st-place" target="_blank">ROHITH INGILELA's notebook</a><br> <a href="https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-place" target="_blank">RSAIDINESHPOLA's notebook</a><br> <a href="https://www.kaggle.com/code/irohith/aslfr-preprocess-dataset" target="_blank">ROHITH INGILELA's pre-process dataset notebook</a></p>
Google - American Sign Language Fingerspelling Recognition	9th Place Solution	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>The below is result of a team effort with <a href="https://www.kaggle.com/rafiko1" target="_blank">@rafiko1</a> and <a href="https://www.kaggle.com/group16" target="_blank">@group16</a> and myself.</p> <p>We want to thank to the hosts for putting together a great competition, it was really fun and we were glad to end up in the gold zone! We've already enjoyed reading some of the top teams solutions- congrats to all who participated.</p> <h2>tl;dr</h2> <p>We used a similar architecture to <a href="https://www.kaggle.com/hoyso48" target="_blank">@hoyso48</a>'s solution from the ASL-sign competition. Our model had six Conv1D/Transformer blocks, was trained with CTC loss over 500+ epochs, and utilized heavy augmentations.</p> <p>This competition posed challenges beyond achieving model accuracy. We had to meet model constraints (5-hour submission limit and 40mb model size). We also needed to make sure all of our code converted nicely to tflite, which at times was frustrating. This required us to find a good balance between model size and inference time with minimal postprocessing.</p> <h2>Data and Preprocessing:</h2> <p>We used the training data as the base dataset, but also trained using the supplemental and external (ChicagoWild/Plus) during fine tuning. Preprocessing was the same for all datasets:</p> <ul> <li>Standard scaling using the mean and standard deviation for each point.</li> <li>x, y, z coordinates for each point.</li> <li>Points used included: RHAND, LHAND, LIP, POSE, REYE, and LEYE.</li> <li>We tried different settings for the max frame length since it had the most impact on our preprocessing and ened up using 356.</li> <li>Downsampling non-hand frames: To handle samples that has more than 356 frames, we first removed frames with missing hands at even intervals. This improved performance when we added it to our inference- so eventually we added it to our training pipeline.</li> <li>Resize: Examples exceeding our max frame length post-downsampling were resized.</li> </ul> <h2>Augmentations:</h2> <p>Heavy augmentations helped us to train long without the risk of overfitting.</p> <ul> <li>Mirror/Flip</li> <li>Random resampling</li> <li>Random rotation</li> <li>Random spatial and temporal masking</li> <li>Temporal cropping</li> <li>Minimal Random noise</li> <li>Random scaling</li> </ul> <h2>Training</h2> <p>Given that training each model took a very long time to train (days running on colab TPUs), we resumed from the best checkpoints to experiment with different augmentations, data ratios, and learning rate schedules. It’s hard to exactly retrace the steps we took to get to our final model but the general idea was:</p> <ul> <li>Base training without augmentations at a consistent learning rate until convergence (~150 epochs).</li> <li>Introducing heavy augmentations and continuing until the validation score plateaued (200-400+ epochs).</li> </ul> <p>Example of how our validation score improved from different experiments over the final week of the competition:<br> <img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2F19aa07e39c404ae7599d66184997d351%2FScreenshot%20from%202023-08-26%2016-58-18.png?generation=1693083542531397&alt=media" alt=""></p> <p>Even up until the end of the competition we were able to get small improvements in our score from more training.</p> <h2>Model Architecture:</h2> <p>Initially, I tried an efficientnet approach similar to the 2nd place in the signs competition. Later, I pivoted to the 1st place architecture similar to public notebooks. After merging teams we found both of us had similar architectures but <a href="https://www.kaggle.com/rafiko1" target="_blank">@rafiko1</a> made some modifications that improved it's performance on this task:</p> <ul> <li>Switched from 3 to 2 Conv1D per Conv/Transformer block.</li> <li>Increated to 6 blocks (resulting in a 10.5M parameter model).</li> <li>Parameters: 384 max frames, 320 dimensions, 8 heads per transformer. </li> <li>Did not use a final MaxPooling layer.</li> </ul> <p>We also had experiments that used DebertaV2 instead of the transformer block, But the final model didn't use them.</p> <h2>Increasing model size without retraining:</h2> <p>Towards the competition's end, realizing we didn't have a great way to ensemble models, we wanted to get the biggest model possible without having to retrain. We found that instead of retraining from scratch we could simply expand the number of Conv1D/Transformer blocks we had - and copy the weights over from our previous best model, repeating the final layer weights. This allowed the model to converge very fast and squeeze out some additional accuracy while saying below the 5 hour inference limit.</p> <h2>Attempts not in our final solution:</h2> <ul> <li><strong>Cutmix based on pseudo labels:</strong> The idea was to use model predictions to pinpoint frames displaying each letter in videos, then applying cutmix at the frame level during training. It functioned initially but became required us to recreate the tfrecords each time we modfiied preprocessing or model archetecture, so it wasn't used in our final solution. When it was used we only applied cutmix across similar phrase types (phone numbers mixed only with phone numbers, etc) but it was interesting to read the top team found it worked best mixing only between the same signer.</li> <li><strong>Ensembling techniques:</strong> Couldn't find a good solution for this with CTC loss in tflite.</li> <li><strong>Beam search</strong></li> <li><strong>Efficientnet/Transformer</strong></li> <li><strong>DebertaV2</strong></li> </ul> <h2>Postprocessing / Ensemble:</h2> <p><strong>Fill word</strong>: As others have noted, there were a significant number of examples in the dataset with very few frames. These had low or no signal in them related to the target. We found it was better to just replace them with a fill word ('+w1- ea-or') We also found that it worked best to apply these to examples that had less than 30 frames AND a prediction phrase that was <= 6 characters.<br> <strong>Model weight averaging</strong> Instead of using the final epoch of each training run, averaging the weights from the final few epochs really helped the CV and LB score. This became less powerful later in the competition when our models were stronger.</p> <h2>Strong Validation/LB Correlation:</h2> <p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F644036%2F2a8e047f8cb571c018f477396c321a91%2Fsub-lb.png?generation=1693083215072772&alt=media" alt=""></p> <p>We tracked each submission's execution time and CV/LB correlation. This competition had a really strong correlation between our validation set (one parquet file) and the LB. Our best private LB score was 0.788 but we didn't select it because it did not perform as well on the public LB.</p>
Google - American Sign Language Fingerspelling Recognition	Custom MovinetClassifier Concept	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p><img src="https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F396181%2F4663191986e89b81a51041daba2605e8%2Fanimation_23842.gif?generation=1692951869820889&alt=media" alt="hand_animation_23842"></p> <p>Above illustrates transforming the hand landmarks (for row 23842 phrase 2 in train) in tensorflow preprocessing layers into a "clip" to be used for inference with a custom classifier fine tuned for the 59 prediction characters.<br> MoViNet models for video classification have small streaming models that can be trained and inferred on CPU and converted to TF Lite. </p> <p><a href="https://github.com/tensorflow/models/tree/master/official/projects/movinet" target="_blank">MoViNet model from TensorFlow Models</a><br> custom classifier MovinetClassifier fine tune with pretrained weights e.g., movinet_a0_stream (trained on Kinetics 600) <br> input shape 1 x 1 x 172 x 172<br> checkpoint ~15 MB trained model ~30 MB TF Lite ~10 MB</p> <p>Convert to TF SavedModel. Then the SavedModel can be converted to TF Lite using the TFLiteConverter<br> <code>converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)</code><br> <code>tflite_model = converter.convert()</code></p> <p><a href="https://arxiv.org/abs/2103.11511" target="_blank">MoViNets: Mobile Video Networks for Efficient Video Recognition</a><br> In the paper, multiple-class labels per video were used for Charades. <br> Or per frame or selected frame inference was a possibility then post process with tensorflow unique_with_counts <br> and considerations for double numbers or letters.</p> <p>This was not finished for the competition, but adding here for interest and awareness of these models. <br> And a bit of fun.<br> It could be a next gen Sign Language uses video content with mediapipe markers and this would fit nicely.</p>
Google - American Sign Language Fingerspelling Recognition	Public notebook, cosmetic changes, 190 position	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Of course, training lasted 50 epochs</p> <p>Hi to all!<br> I haven't been able to really compete in the last while, but I'll share the changes to the public notebook: <a href="https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-" target="_blank">https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-</a> place (special thanks to the author). Only two changes, I reduced the drop_rate and added another Dropout. Here's what the function looks like</p> <p>def get_model(dim = 384,num_blocks = 6,drop_rate = 0.35):<br> inp = tf.keras.Input(INPUT_SHAPE)<br> x = tf.keras.layers.Masking(mask_value=0.0)(inp)<br> x = tf.keras.layers.Dense(dim, use_bias=False, name='stem_conv')(x)<br> pe = tf.cast(positional_encoding(INPUT_SHAPE[0], dim), dtype=x.dtype)<br> x = x + pe<br> x = tf.keras.layers.BatchNormalization(momentum=0.95,name='stem_bn')(x)</p> <pre><code> i range(num_blocks): x = (x) x = (x) x = (x) x = (x) x = tf.keras.layers.(x) x = tf.keras.layers.(x) x = tf.keras.layers.(x) x = tf.keras.layers.,name='classifier')(x) model = tf.keras. loss = CTCLoss # Adam Optimizer optimizer = tfa.optimizers. optimizer = tfa.optimizers. model.compile(loss=loss, optimizer=optimizer) return model </code></pre> <p>The changes are really minor, but worked because the base rate of 0.4 was too big, it slowed learning too much, and I knew from the previous competition that increasing the number of epochs could improve the result. Here, my logic was to simplify learning instead of increasing the number of epochs.<br> Good luck to all of you and a peaceful sky above your head!</p>
Google - American Sign Language Fingerspelling Recognition	Silver - LB 0.770 - Two Lines of Code!	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>That's my guess. Also if we remove missing hand frames, maybe certain consecutive hand activity blends together and fools the model into classifying the wrong letter. These are just guesses, i have not performed EDA nor ablation study to figure out why it helps. But using missing hand frames makes a big difference to the model for some reason.</p> <p>Great trick <a href="https://www.kaggle.com/nightsh4de" target="_blank">@nightsh4de</a> Congratulations on 17th place solo Silver. You were so close to receiving another solo Gold, fantastic!</p> <p>Thanks Ranamalla</p> <p><a href="https://www.kaggle.com/yassinealouini" target="_blank">@yassinealouini</a> Congratulations on becoming Kaggle Discussion Grandmaster!</p> <p>The first thing i did when analyzing the public notebook was to train longer than 50 epochs. That boosted LB from 0.697 to like LB 0.700 with 100 epochs. Then i increased the number of <code>FRAME_LEN</code>. That boost CV and LB too.</p> <p>The next thing i did was analyze the preprocessing code. Lots of time we can improve computer vision, NLP, etc with better preprocess. I saw that the preprocess removed frames, so i tried training a model <strong>without</strong> removing any frames. And wow! The CV and LB boost a lot. Then i changed it to keep 50% frames and that boost CV and LB more and speed up training. In retrospect it makes sense with the CTC loss because the model needs to know how many spacers to add to the output. Providing the missing frames gives the model clues, i think. </p> <p>Next I tried augmentation and training for more epochs. This continued to boost CV and LB. Next i tried simple changes to the architecture but I didn't have much success with that.</p> <p>After reading other solutions, it seems like the way to boost more would be to apply even more aggressive augmentation and train longer.</p> <p>Congratulations <a href="https://www.kaggle.com/goldenlock" target="_blank">@goldenlock</a> Amazing job achieving solo cash gold 3rd place!</p> <p>I didn't think about keeping more non hand frames for the shorter videos (versus long videos), that might help. I did try both keeping 100% and keeping 50%. But i applied my logic to all videos and using 50% did better.</p> <p>Going from 0% non hand to 50% or 100% was a huge boost. I couldn't believe the CV when i first saw it. But i submitted and the LB jumped up just as much.</p> <p>Good question. I'm not sure how much exactly. But removing 50% was better than 0% for both CV and LB. And it made training faster.</p> <p>Thank you Priyanshu1235</p> <p>Thanks Kaggle and Google for hosting a fun ASL competition.</p> <p>I joined a few days ago, so I didn't have time to build my own model. Instead, I began with the best public notebook <a href="https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-place?scriptVersionId=139048726" target="_blank">here</a> (version 17) and attempted to improve it. After making a few changes, I boosted the LB from 0.700 to an amazing 0.770! and obtained Silver medal !!</p> <h1>Change Two Lines of Code</h1> <p>The best public notebook is Rohith Ingilela's awesome public notebook <a href="https://www.kaggle.com/code/irohith/aslfr-ctc-based-on-prev-comp-1st-place" target="_blank">here</a> which was improved by Saidineshpola <a href="https://www.kaggle.com/code/saidineshpola/aslfr-ctc-based-on-prev-comp-1st-place?scriptVersionId=139048726" target="_blank">here</a>. Version 17 of Saidineshpola's notebook achieves CV = 0.689 (scroll to bottom of version 17) and LB = 0.697.</p> <p>That notebook uses TF Records made by Rohith Ingilela <a href="https://www.kaggle.com/code/irohith/aslfr-preprocess-dataset-tfrecords-mean-std" target="_blank">here</a>. An easy trick to boost the performance of the public notebook is create TF Records which keep frames where hands are missing. The TF Records made by linked notebook removes all frames without hands. Instead we can use <code>"output two"</code> below and keep 50% of the frames with missing hands below to boost CV and LB. Updated notebook published <a href="https://www.kaggle.com/code/cdeotte/2-lines-of-code-change-lb-0-760" target="_blank">here</a>.</p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Aug-2023/asl_preprocess.png" alt=""></p> <p>Here is preprocess code to use when making TF Records and during inference:</p> <pre><code>hand = tf.concat(, axis=) hand = tf.where(tf.math.is, , hand) mask = tf.math.not, ) alternating_tensor = tf.math.equal( tf.cumsum( tf.ones ))%, ) mask = tf.math.logical </code></pre> <h1>CTC (Connectionist Temporal Classification) Loss:</h1> <p>In this competition, I learned about CTC loss. This loss is amazing! It allows us to create a model with variable length input and predict variable length output in one step. So we can do things like seq2seq without waiting for a sequential decoder to decode each step. Instead we predict the entire output at once super fast! Giving the model some of the frames with hands missing helps identify duplicates and transistions:</p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Aug-2023/asl_ctc.png" alt=""></p> <h1>Time Augmentation LB +0.004!</h1> <p>The public notebook trains for 50 epochs, I found that training for more epochs continues to boost CV and LB score! My final submission trains for 200 epochs. Furthermore, we can add augmentation (i.e. regularization) which helps the model train longer, prevent overfitting, and generalize better.</p> <p>Below we see the histogram of train data Number of Frames divided by Character Length of Target Phrase. From this plot, we see that the ratio varies a lot. Some videos have a different recorded frame rate than others and some participants sign faster than others.</p> <p><img src="https://raw.githubusercontent.com/cdeotte/Kaggle_Images/main/Aug-2023/frame_hist.png" alt=""></p> <p>What this means is that the model has a hard time transfer learning one frame rate to learn about another frame rate. We can help the model by using time augmentation. For each input sequence, we can randomly shrink frame length by 50% or enlarge 150%. This will add lots of new train data and help the model learn about different frame rates. Here is augmentation code:</p> <pre><code> tf.random.uniform(shape=(), minval=, maxval=)<.: = tf.math.round( tf.random.uniform( =(), minval = tf.cast(tf.shape(lip)[],tf.float32) / ., = tf.cast(tf.shape(lip)[],tf.float32) * .) ) x in</code></pre> <h1>Post Processing LB +0.004!</h1> <p>Sometimes the model doesn't make a good prediction. The shortest train data is length 3. When the model predicts length 2 or less, we know it is a bad prediction. Therefore we can replace bad predictions with the best constant length prediction. Anokas found the best constant length prediction <a href="https://www.kaggle.com/code/anokas/static-greedy-baseline-0-157-lb" target="_blank">here</a>. We add this to our TF Lite mode with the following code:</p> <pre><code>x = tf(tf(x) < , lambda: tf( , tf.int64), lambda: tf(x)) </code></pre> <h1>Solution Code</h1> <p>I published a Kaggle notebook <a href="https://www.kaggle.com/code/cdeotte/2-lines-of-code-change-lb-0-760" target="_blank">here</a> demonstrating the above 3 changes. The first four bullet points below achieve <code>LB = 0.763</code>, then time augmentation boost <code>+0.004</code> and PP boost <code>+0.004</code>. Note that the second bullet point (about batch size) just makes things faster but doesn't change the CV nor LB score. The other bullet points are the key:</p> <ul> <li>Change 2 lines of code to keep 50% missing hand frames</li> <li>Change batch size from 32 to 128 and learning rate from 1e-3 to 4e-3</li> <li>Change FRAME_LEN from 128 to 216</li> <li>Increase epochs 50 to 200</li> <li>Add time augmenation</li> <li>Add post process for preds less than 3 chars</li> </ul> <p>The original notebook resizes the frames to 128. So the original notebook "sees" all the frames. But the original notebook drops the frames without hands (before resize). If we just keep 50% frames without hands and resize to 128, we get a big boost in CV and LB. So in other words even with 128 in both the original and updated, just adding frames without hands (before resize) gives a big boost.</p> <p>Hi <a href="https://www.kaggle.com/ankurlimbashia" target="_blank">@ankurlimbashia</a> As an Nvidia employee, I have access to Nvidia cloud GPUs. So in the past few days, I had a bunch of V100 GPUs running experiments in the cloud. </p> <p>Thanks Gazuwu</p> <p>Thanks Abishek</p> <p>Thank you </p>
Google - American Sign Language Fingerspelling Recognition	Some Notes	Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge	<\|system\|>You can successfully explain and teach others topics and concepts from Kaggle competition solution write-ups. You are a highly talented data scientist who has shown exceptional performance in this data-science Kaggle competition: Google - American Sign Language Fingerspelling Recognition <\|user\|>Explain and teach me concepts from Kaggle competition solution write-ups that solved this challenge <\|assistant\|><p>Many more skilled participants have shared very rich and impressive solutions. I'm only documenting a few experimental insights to avoid adding something redundant to the discussion.</p> <h1>LB History</h1> <table> <thead> <tr> <th></th> <th>public LB</th> <th>private LB</th> </tr> </thead> <tbody> <tr> <td>max_len=188</td> <td>0.770</td> <td>0.741</td> </tr> <tr> <td>max_len=320</td> <td>0.776</td> <td>0.744</td> </tr> <tr> <td>Post Process by Chris Deotte</td> <td>0.778</td> <td>0.747</td> </tr> <tr> <td>expand ratio 2->4</td> <td>0.779</td> <td>0.752</td> </tr> <tr> <td>epochs 126->200</td> <td>0.780</td> <td>0.752</td> </tr> </tbody> </table> <h1>What Didn't Work</h1> <ol> <li><p>I attempted various normalization techniques, including global normalization, local normalization, and normalization centered around the nose. However, I'm unsure if there were any experimental errors, as I found that these approaches merely expedited the training convergence without improving the leaderboard performance. I'm uncertain if my conclusion is accurate.</p></li> <li><p>Lowering the dropout of the final layer from 0.4 to 0.1 and removing dropout from all other layers would result in a decrease of 0.001 points in both the public test set and the private test set.</p></li> <li><p>To place the batch normalization (BN) before the depthwise convolution (DW Conv) instead of after would result in a decrease of 0.001 points for both the public LB (Leaderboard) and private LB (Leaderboard).</p></li> <li><p>Spatial Mask. It may lead to training instability when used. If other aspects are handled well (such as normalization), it might be usable.</p></li> </ol>