[Bug] in evaluation

[LiuShifeng]

I tested the evaluation scripts with the provided codes, config file, and test files.
However, I found the performance is quite different from the human calculation.

I doubt that the parameter name should be labels rather than target_names in semeval2020_0601_eval.py and semeval2020_0602_eval.py.
OR it would be nice that the authors could specify the version of scikit-learn they used.

The script output is shown below and my scikit-learn version is 0.20.3.

              precision    recall  f1-score   support

      HasDef       0.00      0.00      0.00         2
       NoDef       0.60      1.00      0.75         3

   micro avg       0.60      0.60      0.60         5
   macro avg       0.30      0.50      0.37         5
weighted avg       0.36      0.60      0.45         5


              precision    recall  f1-score   support

      B-Term       0.50      0.50      0.50         2
      I-Term       1.00      1.00      1.00         2
B-Definition       0.88      0.78      0.82         9
I-Definition       0.67      0.80      0.73         5

   micro avg       0.78      0.78      0.78        18
   macro avg       0.76      0.77      0.76        18
weighted avg       0.79      0.78      0.78        18


{'Direct-Defines': {'p': 1.0, 'r': 1.0, 'f': 1.0}, 'macro': {'p': 1.0, 'f': 1.0}}