evals.py

import prepare_data
from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix
from seqeval.metrics import classification_report, f1_score, precision_score, recall_score
import matplotlib.pyplot as plt


def compute_metrics(eval_pred):
   y_pred, y_true = prepare_data.align_predictions(eval_pred.predictions, eval_pred.label_ids)
   return {"f1": f1_score(y_true, y_pred), "precision":precision_score(y_true, y_pred), "recall": recall_score(y_true, y_pred)}

def get_f1_score(trainer, dataset):
   return trainer.predict(dataset).metrics["test_f1"]

def get_precision(trainer, dataset):
   return trainer.predict(dataset).metrics["precision"]

def get_recall(trainer, dataset):
   return trainer.predict(dataset).metrics["recall"]

def plot_confusion_matrix(y_preds, y_true, labels, show=False):
   cm = confusion_matrix(y_true, y_preds, normalize="true")
   fig, ax = plt.subplots(figsize=(6, 6))
   disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=labels)
   disp.plot(cmap="Blues", values_format=".2f", ax=ax, colorbar=False)
   plt.title("Normalized confusion matrix")
   if show:
      plt.show()
      
   return cm