File size: 8,406 Bytes
19e161d c3c1db4 fa22072 c3c1db4 082f8b1 c3c1db4 3b101a2 fa22072 19e161d c3c1db4 082f8b1 c3c1db4 3d3da24 438506b c3c1db4 ede5845 ad1aacc ede5845 c3c1db4 ede5845 c3c1db4 ede5845 3d3da24 a6e7a88 ede5845 a6e7a88 3d3da24 438506b c3c1db4 3d3da24 c3c1db4 3d3da24 c3c1db4 747f980 fa22072 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 |
---
tags:
- generated_from_trainer
base_model:
- tartuNLP/EstBERT
metrics:
- accuracy
model-index:
- name: EstBERT128_Rubric
results:
- task:
name: Text Classification
type: text-classification
metrics:
- name: Accuracy
type: accuracy
value: 0.8329238295555115
language: et
license: cc-by-4.0
widget:
- text: >-
Lumesadu ja tuisk levib Kagu-Eestist hommikuks üle maa, päeval läheb sadu
intensiivsemaks. Nähtavus on halb. Lund lisandub 10, kohati kuni 20 cm.
Tiheda saju, tugeva tuule ja tuisu tõttu halvenevad liiklustingimused.
example_title: domestic
- text: >-
Brüsselis puhkenud korruptsiooniskandaalis kahtlustatakse eurosaadikuid
Lähis-Idast meelehea vastuvõtmises. Kinnipeetute seas on üks Euroopa
Parlamendi asepresidente, Belgia prokuratuuri tähelepanu orbiidis teisigi
eurosaadikuid.
example_title: world
- text: >-
Järgmiseks aastaks riigi poolt ette nähtud summa ajakirjanduse
kojukandetoetuseks on sama mis kaks aastat tagasi. See tähendab
märkimisväärset hinnatõusu ja reaalset ohtu, et ajakirjandus on muutumas
luksusteenuseks.
example_title: opinion
pipeline_tag: text-classification
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# EstBERT128_Rubric
This model is a fine-tuned version of [tartuNLP/EstBERT](https://huggingface.co/tartuNLP/EstBERT).
It achieves the following results on the test set:
- Loss: 2.0552
- Accuracy: 0.8329
## How to use?
You can use this model with the Transformers pipeline for text classification.
```
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import pipeline
tokenizer = AutoTokenizer.from_pretrained("tartuNLP/EstBERT128_Rubric")
model = AutoModelForSequenceClassification.from_pretrained("tartuNLP/EstBERT128_Rubric")
nlp = pipeline("text-classification", model=model, tokenizer=tokenizer)
text = "Kaia Kanepi (WTA 57.) langes USA-s Charlestonis toimuval WTA 500 kategooria tenniseturniiril konkurentsist kaheksandikfinaalis, kaotades poolatarile Magda Linette'ile (WTA 64.) 3 : 6, 6 : 4, 2 : 6."
result = nlp(text)
print(result)
```
```
[{'label': 'SPORT', 'score': 0.9999998807907104}]
```
## Model description
A single linear layer classifier is fit on top of the last layer [CLS] token representation of the EstBERT model. The model is fully fine-tuned during training.
## Intended uses & limitations
This model is intended to be used as it is. We hope that it can prove to be useful to somebody but we do not guarantee that the model is useful for anything or that the predictions are accurate on new data.
## Citation information
If you use this model, please cite:
```
@inproceedings{tanvir2021estbert,
title={EstBERT: A Pretrained Language-Specific BERT for Estonian},
author={Tanvir, Hasan and Kittask, Claudia and Eiche, Sandra and Sirts, Kairit},
booktitle={Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa)},
pages={11--19},
year={2021}
}
```
## Training and evaluation data
The model was trained and evaluated on the rubric categories of the [Estonian Valence dataset](http://peeter.eki.ee:5000/valence/paragraphsquery).
The data was split into train/dev/test parts with 70/10/20 proportions.
The nine rubric labels in the Estonian Valence dataset are:
- ARVAMUS (opinion)
- EESTI (domestic)
- ELU-O (life)
- KOMM-O-ELU (comments)
- KOMM-P-EESTI (comments)
- KRIMI (crime)
- KULTUUR (culture)
- SPORT (sports)
- VALISMAA (world)
It probably makes sense to treat the two comments categories (KOMM-O-ELU and KOMM-P-EESTI) as a single category.
## Training procedure
The model was trained for maximu 100 epochs using early stopping procedure. After every epoch, the accuracy was calculated on the development set.
If the development set accuracy did not improve for 20 epochs, the training was stopped.
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 3
- optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-06
- lr_scheduler_type: polynomial
- num_epochs: 100
- mixed_precision_training: Native AMP
### Training results
The final model was taken after 39th epoch.
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:-----:|:---------------:|:--------:|
| 1.1147 | 1.0 | 179 | 0.7421 | 0.7445 |
| 0.4323 | 2.0 | 358 | 0.6863 | 0.7813 |
| 0.1442 | 3.0 | 537 | 0.8545 | 0.7838 |
| 0.0496 | 4.0 | 716 | 1.2872 | 0.7494 |
| 0.0276 | 5.0 | 895 | 1.4702 | 0.7641 |
| 0.0202 | 6.0 | 1074 | 1.3764 | 0.7838 |
| 0.0144 | 7.0 | 1253 | 1.5762 | 0.7887 |
| 0.0078 | 8.0 | 1432 | 1.8806 | 0.7666 |
| 0.0177 | 9.0 | 1611 | 1.6159 | 0.7912 |
| 0.0223 | 10.0 | 1790 | 1.5863 | 0.7936 |
| 0.0108 | 11.0 | 1969 | 1.8051 | 0.7912 |
| 0.0201 | 12.0 | 2148 | 1.9344 | 0.7789 |
| 0.0252 | 13.0 | 2327 | 1.7978 | 0.8084 |
| 0.0104 | 14.0 | 2506 | 1.8779 | 0.7887 |
| 0.0138 | 15.0 | 2685 | 1.6456 | 0.8133 |
| 0.0066 | 16.0 | 2864 | 1.9668 | 0.7912 |
| 0.0148 | 17.0 | 3043 | 2.0068 | 0.7813 |
| 0.0128 | 18.0 | 3222 | 2.1539 | 0.7617 |
| 0.0115 | 19.0 | 3401 | 2.2490 | 0.7838 |
| 0.0186 | 20.0 | 3580 | 2.1768 | 0.7666 |
| 0.0051 | 21.0 | 3759 | 1.8859 | 0.7912 |
| 0.001 | 22.0 | 3938 | 2.0132 | 0.7912 |
| 0.0133 | 23.0 | 4117 | 1.8786 | 0.8084 |
| 0.0149 | 24.0 | 4296 | 2.2307 | 0.7961 |
| 0.014 | 25.0 | 4475 | 2.0041 | 0.8206 |
| 0.0132 | 26.0 | 4654 | 1.8872 | 0.8133 |
| 0.0079 | 27.0 | 4833 | 1.9357 | 0.7961 |
| 0.0078 | 28.0 | 5012 | 2.1891 | 0.7936 |
| 0.0126 | 29.0 | 5191 | 2.0207 | 0.8034 |
| 0.0003 | 30.0 | 5370 | 2.1917 | 0.8010 |
| 0.0015 | 31.0 | 5549 | 2.0417 | 0.8157 |
| 0.0056 | 32.0 | 5728 | 2.1172 | 0.8084 |
| 0.0058 | 33.0 | 5907 | 2.1921 | 0.8206 |
| 0.0001 | 34.0 | 6086 | 2.0079 | 0.8206 |
| 0.0031 | 35.0 | 6265 | 2.2447 | 0.8206 |
| 0.0007 | 36.0 | 6444 | 2.1802 | 0.8084 |
| 0.0061 | 37.0 | 6623 | 2.1103 | 0.8157 |
| 0.0 | 38.0 | 6802 | 2.2265 | 0.8084 |
| 0.0035 | 39.0 | 6981 | 2.0549 | 0.8329 |
| 0.0038 | 40.0 | 7160 | 2.1352 | 0.8182 |
| 0.0001 | 41.0 | 7339 | 2.0975 | 0.8108 |
| 0.0 | 42.0 | 7518 | 2.0833 | 0.8256 |
| 0.0 | 43.0 | 7697 | 2.1020 | 0.8280 |
| 0.0 | 44.0 | 7876 | 2.0841 | 0.8305 |
| 0.0 | 45.0 | 8055 | 2.2085 | 0.8182 |
| 0.0 | 46.0 | 8234 | 2.0756 | 0.8329 |
| 0.0 | 47.0 | 8413 | 2.1237 | 0.8305 |
| 0.0 | 48.0 | 8592 | 2.1217 | 0.8280 |
| 0.0052 | 49.0 | 8771 | 2.3567 | 0.8059 |
| 0.0014 | 50.0 | 8950 | 2.1710 | 0.8206 |
| 0.0032 | 51.0 | 9129 | 2.1452 | 0.8206 |
| 0.0 | 52.0 | 9308 | 2.2820 | 0.8133 |
| 0.0001 | 53.0 | 9487 | 2.2279 | 0.8157 |
| 0.0 | 54.0 | 9666 | 2.1841 | 0.8182 |
| 0.0 | 55.0 | 9845 | 2.1208 | 0.8231 |
| 0.0 | 56.0 | 10024 | 2.0967 | 0.8256 |
| 0.0002 | 57.0 | 10203 | 2.1911 | 0.8231 |
| 0.0 | 58.0 | 10382 | 2.2014 | 0.8231 |
| 0.0 | 59.0 | 10561 | 2.2014 | 0.8182 |
### Framework versions
- Transformers 4.14.1
- Pytorch 1.10.1+cu113
- Datasets 1.16.1
- Tokenizers 0.10.3
### Contact
Kairit Sirts: [email protected] |