INNER CODE UNIT · Python

sentiment_acc

yya518/FinBERT · archive/train_bert.py:84

            sentiment_acc = sentiment_corrects.double() / dataset_sizes[phase]
            assert(len(actual) == len(pred))
            assert(len(actual) == dataset_sizes[phase])
            f1 = f1_score(actual.cpu().numpy(), pred.cpu().numpy(), average='weighted')
            
            print('{} total loss (avg): {:.4f} '.format(phase,epoch_loss ))
            wo.write('{} total loss: {:.4f} \n'.format(phase,epoch_loss ))
            print('{} sentiment_acc: {:.4f}'.format(phase, sentiment_acc))
            wo.write('{} sentiment_acc: {:.4f} \n'.format(phase, sentiment_acc))
            print('{} f1-score: {:.4f}'.format(phase, f1))
            wo.write('{} f1-score:: {:.4f} \n'.format(phase, f1))

            if phase == 'val' and epoch_loss < best_loss:
                print('saving with loss of {}'.format(epoch_loss),
                      'improved over previous {}'.format(best_loss))
                wo.write('saving with loss of {} \n'.format(epoch_loss))
                wo.write('improved over previous {} \n'.format(best_loss))
                wo.write("\n")

View source record →

📰 Research Paper
Loading…
⏳ Fetching content…