Datasets:

---

calvpang

/

github-issues

like 0

[]

[]

[ "This library uses Apache Arrow under the hood to store datasets on disk.\r\nThe advantage of Apache Arrow is that it allows to memory map the dataset. This allows to load datasets bigger than memory and with almost no RAM usage. It also offers excellent I/O speed.\r\n\r\nFor example when you access one element or one batch\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nsquad = load_dataset(\"squad\", split=\"train\")\r\nfirst_element = squad[0]\r\none_batch = squad[:8]\r\n```\r\n\r\nthen only this element/batch is loaded in memory, while the rest of the dataset is memory mapped.", "How can we change how much data is loaded to memory with Arrow? I think that I am having some performance issue with it. When Arrow loads the data from disk it does it in multiprocess? It's almost twice slower training with arrow than in memory.\r\n\r\nEDIT:\r\nMy fault! I had not seen the `dataloader_num_workers` in `TrainingArguments` ! Now I can parallelize and go fast! Sorry, and thanks.", "> How can we change how much data is loaded to memory with Arrow? I think that I am having some performance issue with it. When Arrow loads the data from disk it does it in multiprocess? It's almost twice slower training with arrow than in memory.\r\n\r\nLoading arrow data from disk is done with memory-mapping. This allows to load huge datasets without filling your RAM.\r\nMemory mapping is almost instantaneous and is done within one process.\r\n\r\nThen, the speed of querying examples from the dataset is I/O bounded depending on your disk. If it's an SSD then fetching examples from the dataset will be very fast.\r\nBut since the I/O speed of an SSD is lower than the one of RAM it's expected to be slower to fetch data from disk than from memory.\r\nStill, if you load the dataset in different processes then it can be faster but there will still be the I/O bottleneck of the disk.\r\n\r\n> EDIT:\r\n> My fault! I had not seen the `dataloader_num_workers` in `TrainingArguments` ! Now I can parallelize and go fast! Sorry, and thanks.\r\n\r\nOk let me know if that helps !\r\n" ]

[]

[]

[]

[]

[]

[ "Could you fix the dummy data before we merge ?\r\nLooks like the dummy `train.csv` is missing", "Thanks @Narsil @lhoestq for adding MeDAL :)" ]

[]

[]

[ "Dupe" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "#978 is working on adding code refinement\r\n\r\nmaybe we should keep the CodeXGlue benchmark (as glue) and don't merge the code_refinement dataset proposed in #978 ?\r\n\r\ncc @reshinthadithyan", "Hi @madlag and @lhoestq , I am extremely interested in getting this dataset into HF's library as I research in this area a lot. I see that it hasn't been updated in a while, but it is very close to being finished. If no one is currently working on this, I'd be happy to do any final touches that might be needed to get this merged.", "Hi @ncoop57 ! Thanks for your interest and sorry for the inactivity on this PR.\r\nSure feel free to create another PR to continue this one ! This one was really close to being merged so I think it won't require that much changes. In addition to my previous comments, there should also be a \"Contributions\" subsection (see the template of the README [here](https://github.com/huggingface/datasets/blob/master/templates/README.md))", "Superseded by https://github.com/huggingface/datasets/pull/2357 ." ]

[]

[]

[ "Looks like the google drive download failed.\r\nI'm getting a `Google Drive - Quota exceeded` error while looking at the downloaded file.\r\n\r\nWe should consider finding a better host than google drive for this dataset imo\r\nrelated : #873 #864 ", "It is working now, thank you. \r\n\r\nShould I leave this issue open to address the Quota-exceeded error?", "Yes please. It's been happening several times, we definitely need to address it", "Any updates on this one? I'm facing a similar issue trying to add CelebA.", "I've looked into it and couldn't find a solution. This looks like a Google Drive limitation..\r\nPlease try to use other hosts when possible", "The original links are google drive links. Would it be feasible for HF to maintain their own servers for this? Also, I think the same issue must also exist with TFDS.", "It's possible to host data on our side but we should ask the authors. TFDS has the same issue and doesn't have a solution either afaik.\r\nOtherwise you can use the google drive link, but it it's not that convenient because of this quota issue.", "Okay. I imagine asking every author who shares their dataset on Google Drive will also be cumbersome.", "I am getting this error as well. Is there a fix?", "Not as long as the data is stored on GG drive unfortunately.\r\nMaybe we can ask if there's a mirror ?\r\n\r\nHi @JafferWilson is there a download link to get cnn dailymail from another host than GG drive ?\r\n\r\nTo give you some context, this library provides tools to download and process datasets. For CNN DailyMail the data are downloaded from the link you provide on your github repository. Unfortunately because of GG drive quotas, many users are not able to load this dataset.", "The following copy of CNN/DM dataset, fixed the problem for me:\r\nhttps://huggingface.co/datasets/ccdv/cnn_dailymail", "Thanks for the link @mrazizi !\r\n\r\nApparently the original authors don't host the dataset themselves (\"for legal reasons\", source [here](https://github.com/abisee/cnn-dailymail/issues/9))." ]

[]

[]

[ "Blocked @k125-ak ;) Bye-bye" ]

[]

[]

[ "Looks like the PR includes commits about many other files.\r\nCould you create a clean branch from master, and create another PR ?", "Sorry, will do that. " ]

[]

[]

[ "Hi @hfawaz ! This is working fine for me. Is it a repeated occurence? Have you tried from the latest verion?", "Hi, it seems a connection problem. \r\nNow it says: \r\n`ConnectionError: Couldn't reach https://amazon-reviews-ml.s3-us-west-2.amazonaws.com/json/train/dataset_ja_train.json`" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "my implementation of sharding per TPU core: https://github.com/google-research/ruse/blob/d4dd58a2d8efe0ffb1a9e9e77e3228d6824d3c3c/seq2seq/trainers/t5_trainer.py#L316 \r\nmy implementation of dataloader for this case https://github.com/google-research/ruse/blob/d4dd58a2d8efe0ffb1a9e9e77e3228d6824d3c3c/seq2seq/tasks/tasks.py#L496 " ]

[]

[]

[ "merging since the CI is fixed on master" ]

[]

[]

[ "CI failures seem to be unrelated (related to `norwegian_ner`)\r\n```\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_class_norwegian_ner\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_configs_norwegian_ner\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_load_dataset_norwegian_ner\r\n```", "you can just rebase from master to fix the CI :) ", "can you just rebase from master before we merge ?", "Sorry, the rebase from master went horribly wrong, I guess I'll just open another PR\r\n\r\nClosing this one due to a mistake in rebasing :(", "Continued in #1014 " ]

[]

[]

[ "This dataset already exists apparently, sorry :/ \r\nsee\r\nhttps://github.com/huggingface/datasets/blob/master/datasets/gap/gap.py\r\n\r\nFeel free to re-use the dataset card you did for `/datasets/gap`\r\n", "oh heck, my bad 🤦‍♂️ sorry" ]

[]

[]

[ "can't find the Whoa file since there' nothing left", "The classic `rm -rf` command - nice one" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "Thank you for contributing a Thai dataset, @mapmeld ! I'm super hyped. \r\nOne comment I may add is that wongnai-corpus has two datasets: review classification (this) and word tokenization (https://github.com/wongnai/wongnai-corpus/blob/master/search/labeled_queries_by_judges.txt).\r\nWould it be possible for you to rename this one something along the line of `wongnai-reviews` so that when/if we include the word tokenization dataset, we will know which is which.\r\n\r\nThis helps solve my check_code_quality issue.\r\n```\r\nmake style\r\nblack --line-length 119 --target-version py36 datasets/wongnai\r\nflake8 datasets/wongnai\r\nisort datasets/wongnai/wongnai.py\r\n```", "@cstorm125 thanks! following your suggestions on formatting and on naming the dataset\r\n\r\nI am writing a blog post about Thai NLP and transformers (example: mBERT does 1-2 character tokens instead of doing word segmentation), started adding this dataset to use as an example, and then saw you were adding other datasets. Great work! And if you know any Thai BERT models beyond https://github.com/ThAIKeras/bert we should maybe talk over email!" ]

[]

[]

[]

[]

[]

[ "Also cc @madlag since I recall you wanted to work on CodeXGlue as well ?", "Yes, sorry I did not see earlier your message. I added 34 on the 35 datasets in CodeXGlue, tomorrow I will wrap it up, and so I will remove my version for code_refinement. Maybe we can just have a renaming after the merge, to have a consistent naming with all the other codexglue datasets ? What do you think @reshinthadithyan ?", "> Yes, sorry I did not see earlier your message. I added 34 on the 35 datasets in CodeXGlue, tomorrow I will wrap it up, and so I will remove my version for code_refinement. Maybe we can just have a renaming after the merge, to have a consistent naming with all the other codexglue datasets ? What do you think @reshinthadithyan ?\r\n\r\nHello @madlag, I think you can retain that in your script. Let's stick onto the same file like how Glue is maintained.", "Hi @reshinthadithyan ! Are you still working on this version of the dataset or are we going with @madlag 's only ?", "> Hi @reshinthadithyan ! Are you still working on this version of the dataset or are we going with @madlag 's only ?\r\n\r\nHello, yes. We are going with Madlag's" ]

[]

[]

[]

[]

[]

[ "looks like this PR includes changes about many other files than the oens for Araboc POS Dialect\r\n\r\nCan you create a another branch and another PR please ?", "Sorry! I'm not sure how I managed to do that. I'll make a new branch." ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "I have to manually copy a dataset_infos.json file from other dataset and modify it since the `datasets-cli` isn't able to handle manually downloaded datasets yet (as far as I know).", "you can generate the dataset_infos.json file even for dataset with manual data\r\nTo do so just specify `--data_dir <path/to/the/folder/containing/the/manual/data>`", "Also, dummy_data seems having difficulty to handle manually downloaded datasets. `python datasets-cli dummy_data datasets/ms_terms --data_dir ...` reported `error: unrecognized arguments: --data_dir` error. Without `--data_dir`, it reported this error:\r\n```\r\nDataset ms_terms with config BuilderConfig(name='ms_terms-full', version=1.0.0, data_dir=None, data_files=None, description='...\\n') seems to already open files in the method `_split_generators(...)`. You might consider to instead only open files in the method `_generate_examples(...)` instead. If this is not possible the dummy data has to be created with less guidance. Make sure you create the file None.\r\nTraceback (most recent call last):\r\n File \"datasets-cli\", line 36, in <module>\r\n service.run()\r\n File \"/Users/lzhao/Downloads/huggingface/datasets/src/datasets/commands/dummy_data.py\", line 326, in run\r\n dataset_builder=dataset_builder, mock_dl_manager=mock_dl_manager\r\n File \"/Users/lzhao/Downloads/huggingface/datasets/src/datasets/commands/dummy_data.py\", line 406, in _print_dummy_data_instructions\r\n for split in generator_splits:\r\nUnboundLocalError: local variable 'generator_splits' referenced before assignment\r\n```", "Oh yes `--data_dir` seems to only be supported for the `datasets_cli test` command. Sorry about that.\r\n\r\nCan you try to build the dummy_data.zip file manually ?\r\n\r\nIt has to be inside `./datasets/ms_terms/dummy/ms_terms-full/1.0.0`.\r\nInside this folder, please create a folder `dummy_data` that contains a dummy file `MicrosoftTermCollection.tbx` (with just a few examples in it). Then you can zip the `dummy_data` folder to `dummy_data.zip`\r\n\r\nThen you can check if it worked using the command\r\n```\r\npytest tests/test_dataset_common.py::LocalDatasetTest::test_load_dataset_ms_terms\r\n```\r\n\r\nFeel free to use some debugging print statements in your script if it doesn't work first try to see what `dl_manager.manual_dir` ends up being and also `path_to_manual_file`.\r\n\r\nFeel free to ping me if you have other questions", "`pytest tests/test_dataset_common.py::LocalDatasetTest::test_load_dataset_ms_terms` gave `1 passed, 4 warnings in 8.13s`. Existing datasets, like `wikihow`, and `newsroom`, also report 4 warnings. So, I guess that is not related to this dataset.", "Could you run `make style` before we merge @leoxzhao ?", "the other errors are fixed on master so it's fine", "> Could you run `make style` before we merge @leoxzhao ?\r\n\r\nSure thing. Done. Thanks Quentin. I have other datasets in mind. All of which requires manual download. This process is very helpful", "Thank you :) " ]

[]

[]

[ "Hi @lhoestq,\r\n\r\nI guess this PR can be closed since we merged #2044?\r\n\r\nI have used the same link for the homepage, as it is where the dataset is provided, hope that is okay?", "Closing in favor of #2044, thanks again :)\r\n\r\n> I have used the same link for the homepage, as it is where the dataset is provided, hope that is okay?\r\n\r\nYea it's ok actually, at that time I thought there was another homepage for this dataset" ]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[]

[ "One trick if you want to add other datasets: consider running these commands each time you want to add a new dataset\r\n```\r\ngit checkout master\r\ngit fetch upstream\r\ngit rebase upstream/master\r\ngit checkout -b add-<my_dataset_name>\r\n```" ]

[]

[]

[ "Oh yeah, for some reason I thought you had to do it after the merge, I'll get on it", "Weird, now the CI seems to fail because of other datasets (XGLUE, Norwegian_NER)", "Yea you just need to rebase from master", "Re-opening a PR without the messed-up rebase" ]

[]

[]

[ "Looks like your PR now shows changes in many other files than the ones for CLINC150.\r\nFeel free to create another branch and another PR", "created new [PR](https://github.com/huggingface/datasets/pull/1016)\r\n\r\nclosing this!" ]

[]

[]

[]

[]

[]

[ "This is task is part of the GEM suite so will actually need a more complete dataset card. I'm taking a break for now though and will get back to it before merging :) " ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "here I share my dataloader currently for multiple tasks: https://gist.github.com/rabeehkarimimahabadi/39f9444a4fb6f53dcc4fca5d73bf8195 \r\n\r\nI need to train my model distributedly with this dataloader, \"MultiTasksataloader\", currently this does not work in distributed fasion,\r\nto save on memory I tried to use iterative datasets, could you have a look in this dataloader and tell me if this is indeed the case? not sure how to make datasets being iterative to not load them in memory, then I remove the sampler for dataloader, and shard the data per core, could you tell me please how I should implement this case in datasets library? and how do you find my implementation in terms of correctness? thanks \r\n" ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "@lhoestq I added the README files, and now the tests fail... (check commit history, only changed MD file)\r\nThe tests seem a bit unstable", "the `RemoteDatasetTest ` errors in the CI are fixed on master so it's fine", "merging since the CI is fixed on master" ]

[]

[]

[]

[]

[]

[ "Merging this one, good job and thank you @jplu :) " ]

[]

[]

[ "> Really cool ! Thanks for adding this one :)\r\n> Good job at adding all those citations for each task\r\n> \r\n> Looks like the dummy data test doesn't pass. Maybe some files are missing in the dummy_data.zip files ?\r\n> The error reports `pragmeval/verifiability/train.tsv` to be missing\r\n> \r\n> Also could you add the tags part of the dataset card (the rest is optional) ?\r\n> See more info here : https://github.com/huggingface/datasets/blob/master/ADD_NEW_DATASET.md#tag-the-dataset-and-write-the-dataset-card\r\n\r\nIn the prior commits I generated dataset_infos and the dummy files myself\r\nNow they are generated with the cli, and the tests now seem to be passing better\r\nI will look into the tag\r\n", "Looks like you did a good job with dummy data in the first place !\r\nThe downside of automatically generated dummy data is that the files are heavier (here 40KB per file).\r\nIf you could replace the generated dummy files with the one you created yourself it would be awesome, since the one you did yourself are way lighter (around 1KB per file). Using small files make `git clone` run faster so we encourage to use small dummy_data files.", "could you rebase from master ? it should fix the CI", "> could you rebase from master ? it should fix the CI\r\n\r\nI think it is due to the file structure of the dummy data that causes test failure. The automatically generated dummy data pass the tests", "Indeed the error reports that `pragmeval/verifiability/train.tsv` is missing for the verifiability dummy_data.zip file.\r\nTo fix that you should add the missing data files in each dummy_data.zip file.\r\nTo test that your dummy data work you can run\r\n```\r\nRUN_SLOW=1 pytest tests/test_dataset_common.py::LocalDatasetTest::test_load_dataset_all_configs_\r\n```\r\nif some file is missing it should tell you which one", "Also it looks like you haven't rebased from master yet, even though you did a `rebase` commit. \r\n\r\nrebasing should fix the other CI fails", "It's ok if we have `RemoteDatasetTest ` errors, they're fixed on master", "merging since the CI is fixed on master", "Hey @sileod! Super nice to see you participating ;)\r\n\r\nDid you officially joined the sprint by posting on [the forum thread](https://discuss.huggingface.co/t/open-to-the-community-one-week-team-effort-to-reach-v2-0-of-hf-datasets-library/2176) and joining our slack?\r\n\r\nI can't seem to find you there! Should I add you directly with your gmail address?", "Hi @sileod 👋 " ]

[]

[]

[ "Test failing for same issues as https://github.com/huggingface/datasets/pull/939\r\nPlease advise.\r\n\r\n```\r\n=========================== short test summary info ============================\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_class_flue\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_class_norwegian_ner\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_configs_flue\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_builder_configs_norwegian_ner\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_load_dataset_flue\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_load_dataset_norwegian_ner\r\nFAILED tests/test_dataset_common.py::RemoteDatasetTest::test_load_dataset_xglue\r\n===== 7 failed, 1309 passed, 932 skipped, 11 warnings in 166.71s (0:02:46) =====\r\n```", "Closing and opening a new pull request to solve rebase issues", "To be continued on https://github.com/huggingface/datasets/pull/982" ]

[]

[]

[ "Hi @lhoestq,\r\nInitially I tried int(-1) only in place of nan labels and missing values but I kept on getting this error ```pyarrow.lib.ArrowTypeError: Expected bytes, got a 'int' object``` maybe because I'm sending int values (-1) to objects which are string type" ]

[]

[]

[]

[]

[]

[ "Wrongly branching from existing branch of wisesight_sentiment. Closing and opening another one specifically for prachathai67k" ]

[]

[]

[]

[]

[]

[ "@lhoestq added. " ]

[]

[]

[]

[]

[]

[]

[]

[]

[ "The checks failed again even if I didn't make any changes.", "you just need to rebase from master to fix the CI :)", "Sorry for the mess, I'm confused by the rebase and thus created a new branch." ]

[]

[]

[ "Replaced by #1126" ]

[]

[]

[ "thanks ! merging this one" ]