Hugging Face's logo

saattrupdan
/
verdict-classifier-en

Text Classification
Transformers
PyTorch
TensorBoard
Safetensors
English
roberta
Generated from Trainer
Model card Files Metrics Community
1
verdict-classifier-en / README.md
saattrupdan's picture
saattrupdan
Update README.md
458e8ed
preview code
|
raw
history blame
6.46 kB
metadata
license: mit
language: en
tags:
  - generated_from_trainer
model-index:
  - name: verdict-classifier-en
    results:
      - task:
          type: text-classification
          name: Verdict Classification
widget:
  - One might think that this is true, but it's taken out of context.

English Verdict Classifier

This model is a fine-tuned version of roberta-base on 2,500 deduplicated verdicts from Google Fact Check Tools API, translated into English with the Google Cloud Translation API. It achieves the following results on the evaluation set, being 1,000 such verdicts translated into English, but here including duplicates to represent the true distribution:

  • Loss: 0.1304
  • F1 Macro: 0.8868
  • F1 Misinformation: 0.9832
  • F1 Factual: 0.9890
  • F1 Other: 0.6882
  • Prec Macro: 0.8580
  • Prec Misinformation: 0.9918
  • Prec Factual: 0.9783
  • Prec Other: 0.6038

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 625
  • num_epochs: 1000

Training results

Training Loss Epoch Step Validation Loss F1 Macro F1 Misinformation F1 Factual F1 Other Prec Macro Prec Misinformation Prec Factual Prec Other
1.0588 0.64 50 1.0803 0.0256 0.0 0.0 0.0768 0.0133 0.0 0.0 0.0400
0.9885 1.28 100 1.0055 0.3497 0.9291 0.0 0.12 0.3910 0.8729 0.0 0.3
0.971 1.92 150 0.9218 0.3102 0.9306 0.0 0.0 0.2900 0.8701 0.0 0.0
0.9263 2.56 200 0.6035 0.3102 0.9306 0.0 0.0 0.2900 0.8701 0.0 0.0
0.8672 3.2 250 0.3639 0.4428 0.9337 0.0 0.3946 0.3976 0.9217 0.0 0.2710
0.743 3.84 300 0.2396 0.7944 0.9698 0.9091 0.5043 0.7893 0.9812 1.0 0.3867
0.5106 4.49 350 0.1579 0.8399 0.9733 0.9888 0.5577 0.8130 0.9859 1.0 0.4531
0.4215 5.13 400 0.1245 0.8174 0.9747 0.9834 0.4941 0.8076 0.9780 0.9780 0.4667
0.3941 5.77 450 0.1422 0.8298 0.9678 1.0 0.5217 0.7960 0.9880 1.0 0.4
0.3105 6.41 500 0.1352 0.8223 0.9696 0.9836 0.5138 0.7872 0.9881 0.9677 0.4058
0.3126 7.05 550 0.1126 0.8423 0.9756 0.9945 0.5567 0.8162 0.9859 0.9890 0.4737
0.2206 7.69 600 0.1206 0.8557 0.9761 0.9890 0.6019 0.8203 0.9905 0.9783 0.4921
0.2472 8.33 650 0.1296 0.8481 0.9731 0.9945 0.5766 0.8105 0.9917 0.9890 0.4507
0.1839 8.97 700 0.1357 0.8582 0.9761 0.9890 0.6095 0.8208 0.9917 0.9783 0.4923
0.1282 9.61 750 0.1465 0.8481 0.9756 0.9945 0.5743 0.8175 0.9882 0.9890 0.4754
0.1447 10.26 800 0.1621 0.8602 0.9767 0.9945 0.6095 0.8243 0.9917 0.9890 0.4923
0.1223 10.9 850 0.1304 0.8868 0.9832 0.9890 0.6882 0.8580 0.9918 0.9783 0.6038
0.1053 11.54 900 0.1640 0.8714 0.9797 0.9945 0.64 0.8380 0.9918 0.9890 0.5333
0.064 12.18 950 0.1983 0.8627 0.9791 0.9889 0.62 0.8321 0.9906 0.9889 0.5167
0.1085 12.82 1000 0.1811 0.8688 0.9803 0.9945 0.6316 0.8413 0.9895 0.9890 0.5455
0.0885 13.46 1050 0.2052 0.8710 0.9821 0.9945 0.6364 0.8532 0.9872 0.9890 0.5833
0.0799 14.1 1100 0.1826 0.8801 0.9827 0.9836 0.6742 0.8565 0.9895 0.9677 0.6122
0.0737 14.74 1150 0.2158 0.8556 0.9761 0.9945 0.5962 0.8213 0.9905 0.9890 0.4844
0.0564 15.38 1200 0.2283 0.8637 0.9797 0.9945 0.6170 0.8381 0.9883 0.9890 0.5370
0.0547 16.03 1250 0.2508 0.8693 0.9785 0.9888 0.6408 0.8381 0.9906 1.0 0.5238
0.0602 16.67 1300 0.2320 0.8555 0.9798 0.9889 0.5977 0.8420 0.9838 0.9889 0.5532
0.0576 17.31 1350 0.2346 0.8737 0.9803 0.9945 0.6465 0.8411 0.9918 0.9890 0.5424

Framework versions

  • Transformers 4.11.3
  • Pytorch 1.9.0+cu102
  • Datasets 1.9.0
  • Tokenizers 0.10.2