15b19130a9162feb9153a2f38c5c8b0af02c6a1d,train.py,,eval_epoch,#,75

Before Change

        n_words = gold.data.ne(Constants.PAD).sum()
        n_total_words += n_words
        n_total_correct += n_correct
        total_loss += loss.data[0]

    return total_loss/n_total_words, n_total_correct/n_total_words

def train(model, training_data, validation_data, crit, optimizer, opt):

After Change

    n_total_words = 0
    n_total_correct = 0

    with torch.no_grad():
        for batch in tqdm(
                validation_data, mininterval=2,
                desc="  - (Validation) ", leave=False):

            // prepare data
            src_seq, src_pos, tgt_seq, tgt_pos = map(lambda x: x.to(device), batch)
            gold = tgt_seq[:, 1:]

            // forward
            pred = model(src_seq, src_pos, tgt_seq, tgt_pos)
            loss, n_correct = get_performance(crit, pred, gold)

            // note keeping
            n_words = gold.data.ne(Constants.PAD).sum()
            n_total_words += n_words
            n_total_correct += n_correct
            total_loss += loss.item()

    loss_per_word = total_loss/n_total_words
    accuracy = n_total_correct/n_total_words
    return loss_per_word, accuracy

Instances