Replace seqeval code with original classification_report for simplicity

[Hironsan]

Recently, the original seqeval has enabled us to get per type scores and overall scores as a dictionary.

This PR replaces the current code with the original function(classification_report) to simplify it.

Also, the original code has been updated to fix #352.

• Related issue: 🐛[BugFix]fix get enitity chakki-works/seqeval#38

from datasets import load_metric
metric = load_metric("seqeval")
y_true = [['O', 'O', 'O', 'B-MISC', 'I-MISC', 'I-MISC', 'O'], ['B-PER', 'I-PER', 'O']]
y_pred = [['O', 'O', 'B-MISC', 'I-MISC', 'I-MISC', 'I-MISC', 'O'], ['B-PER', 'I-PER', 'O']]
metric.compute(predictions=y_pred, references=y_true)
# Output: {'MISC': {'precision': 0.0, 'recall': 0.0, 'f1': 0, 'number': 1}, 'PER': {'precision': 1.0, 'recall': 1.0, 'f1': 1.0, 'number': 1}, 'overall_precision': 0.5, 'overall_recall': 0.5, 'overall_f1': 0.5, 'overall_accuracy': 0.8}

[thomwolf]

LGTM, thanks!

[lhoestq]

That's awesome thank you :)

Could you update the docstring before we merge the PR ?

[lhoestq]

Thank you !

[Tomas0413]

Hello,

I ran https://github.com/huggingface/transformers/blob/master/examples/token-classification/run.sh

And received this error:

100%|██████████| 407/407 [21:37<00:00,  3.44s/it]Traceback (most recent call last):
  File "run_ner.py", line 445, in <module>
    main()
  File "run_ner.py", line 398, in main
    results = trainer.evaluate()
  File "/data/2021/transformers/src/transformers/trainer.py", line 1470, in evaluate
    metric_key_prefix=metric_key_prefix,
  File "/data/2021/transformers/src/transformers/trainer.py", line 1622, in prediction_loop
    metrics = self.compute_metrics(EvalPrediction(predictions=preds, label_ids=label_ids))
  File "run_ner.py", line 345, in compute_metrics
    results = metric.compute(predictions=true_predictions, references=true_labels)
  File "/usr/local/lib/python3.6/dist-packages/datasets/metric.py", line 398, in compute
    output = self._compute(predictions=predictions, references=references, **kwargs)
  File "/root/.cache/huggingface/modules/datasets_modules/metrics/seqeval/81eda1ff004361d4fa48754a446ec69bb7aa9cf4d14c7215f407d1475941c5ff/seqeval.py", line 97, in _compute
    report = classification_report(y_true=references, y_pred=predictions, suffix=suffix, output_dict=True)
TypeError: classification_report() got an unexpected keyword argument 'output_dict'

I'm still trying multiple things to see if I can work around this, but I thought it might be useful to mention it here.

Name: transformers
Version: 4.3.0.dev0

Name: datasets
Version: 1.2.1

[lhoestq]

Hi, can you try to update your local installation of seqeval ?

pip install --upgrade seqeval

[Tomas0413]

@lhoestq thanks for the reply. Indeed it was some issue with my setup. I removed the "transformers" and "datasets" (that I had previously installed from the source code), cleared the cache and installed everything again. It works great now!