baseline_model/data_utils/train.py [34:98]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def cal_performance(pred, gold, trg_pad_idx, smoothing=False):
    ''' Apply label smoothing if needed '''

    loss = cal_loss(pred, gold, trg_pad_idx, smoothing=smoothing)

    pred = pred.max(1)[1]
    gold = gold.contiguous().view(-1)
    non_pad_mask = gold.ne(trg_pad_idx)
    n_correct = pred.eq(gold).masked_select(non_pad_mask).sum().item()
    n_word = non_pad_mask.sum().item()

    return loss, n_correct, n_word


def cal_loss(pred, gold, trg_pad_idx, smoothing=False):
    ''' Calculate cross entropy loss, apply label smoothing if needed. '''

    gold = gold.contiguous().view(-1)

    if smoothing:
        eps = 0.1
        n_class = pred.size(1)

        one_hot = torch.zeros_like(pred).scatter(1, gold.view(-1, 1), 1)
        one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
        log_prb = F.log_softmax(pred, dim=1)

        non_pad_mask = gold.ne(trg_pad_idx)
        loss = -(one_hot * log_prb).sum(dim=1)
        loss = loss.masked_select(non_pad_mask).sum()  # average later
    else:
        loss = F.cross_entropy(pred, gold, ignore_index=trg_pad_idx, reduction='sum')
        # loss = F.cross_entropy(pred, gold, ignore_index = trg_pad_idx)
        # pdb.set_trace()
        # loss = F.cross_entropy(pred, gold, ignore_index=trg_pad_idx, reduction='sum')
    return loss

def epoch_time(start_time, end_time):
    elapsed_time = end_time - start_time
    elapsed_mins = int(elapsed_time / 60)
    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))
    return elapsed_mins, elapsed_secs

def tokenize_de(text):
    """
    Tokenizes German text from a string into a list of strings (tokens) and reverses it
    """
    return [tok.text for tok in spacy_de.tokenizer(text)][::-1]

def tokenize_en(text):
    """
    Tokenizes English text from a string into a list of strings (tokens)
    """
    return [tok.text for tok in spacy_en.tokenizer(text)]

def patch_trg(trg, pad_idx):
    trg = trg.transpose(0, 1)
    trg, gold = trg[:, :-1], trg[:, 1:].contiguous().view(-1)
    return trg, gold

def print_performances(header, loss, accu, start_time):
    print('  - {header:12} ppl: {ppl: 8.5f}, accuracy: {accu:3.3f} %, '\
          'elapse: {elapse:3.3f} min'.format(
              header=f"({header})", ppl=math.exp(min(loss, 100)),
              accu=100*accu, elapse=(time.time()-start_time)/60))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



baseline_model/data_utils/train_vul.py [74:137]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def cal_performance(pred, gold, trg_pad_idx, smoothing=False):
    ''' Apply label smoothing if needed '''

    loss = cal_loss(pred, gold, trg_pad_idx, smoothing=smoothing)

    pred = pred.max(1)[1]
    gold = gold.contiguous().view(-1)
    non_pad_mask = gold.ne(trg_pad_idx)
    n_correct = pred.eq(gold).masked_select(non_pad_mask).sum().item()
    n_word = non_pad_mask.sum().item()

    return loss, n_correct, n_word

def cal_loss(pred, gold, trg_pad_idx, smoothing=False):
    ''' Calculate cross entropy loss, apply label smoothing if needed. '''

    gold = gold.contiguous().view(-1)

    if smoothing:
        eps = 0.1
        n_class = pred.size(1)

        one_hot = torch.zeros_like(pred).scatter(1, gold.view(-1, 1), 1)
        one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
        log_prb = F.log_softmax(pred, dim=1)

        non_pad_mask = gold.ne(trg_pad_idx)
        loss = -(one_hot * log_prb).sum(dim=1)
        loss = loss.masked_select(non_pad_mask).sum()  # average later
    else:
        loss = F.cross_entropy(pred, gold, ignore_index=trg_pad_idx, reduction='sum')
        # loss = F.cross_entropy(pred, gold, ignore_index = trg_pad_idx)
        # pdb.set_trace()
        # loss = F.cross_entropy(pred, gold, ignore_index=trg_pad_idx, reduction='sum')
    return loss

def epoch_time(start_time, end_time):
    elapsed_time = end_time - start_time
    elapsed_mins = int(elapsed_time / 60)
    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))
    return elapsed_mins, elapsed_secs

def tokenize_de(text):
    """
    Tokenizes German text from a string into a list of strings (tokens) and reverses it
    """
    return [tok.text for tok in spacy_de.tokenizer(text)][::-1]

def tokenize_en(text):
    """
    Tokenizes English text from a string into a list of strings (tokens)
    """
    return [tok.text for tok in spacy_en.tokenizer(text)]

def patch_trg(trg, pad_idx):
    trg = trg.transpose(0, 1)
    trg, gold = trg[:, :-1], trg[:, 1:].contiguous().view(-1)
    return trg, gold

def print_performances(header, loss, accu, start_time):
    print('  - {header:12} ppl: {ppl: 8.5f}, accuracy: {accu:3.3f} %, '\
          'elapse: {elapse:3.3f} min'.format(
              header=f"({header})", ppl=math.exp(min(loss, 100)),
              accu=100*accu, elapse=(time.time()-start_time)/60))
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -