Labels smoothing and categorical loss functions - alternatives?

Brando_Miranda · March 23, 2021, 8:11pm

is the answer from here good? @ptrblck I am sort of confused what I should be using, to many options nothing that seems official pytorch.

github.com

OpenNMT/OpenNMT-py/blob/e8622eb5c6117269bb3accd8eb6f66282b5e67d9/onmt/utils/loss.py#L186


        num_non_padding = non_padding.sum().item()
        return onmt.utils.Statistics(loss.item(), num_non_padding, num_correct)

    def _bottle(self, _v):
        return _v.view(-1, _v.size(2))

    def _unbottle(self, _v, batch_size):
        return _v.view(-1, batch_size, _v.size(1))


class LabelSmoothingLoss(nn.Module):
    """
    With label smoothing,
    KL-divergence between q_{smoothed ground truth prob.}(w)
    and p_{prob. computed by model}(w) is minimized.
    """
    def __init__(self, label_smoothing, tgt_vocab_size, ignore_index=-100):
        assert 0.0 < label_smoothing <= 1.0
        self.ignore_index = ignore_index
        super(LabelSmoothingLoss, self).__init__()

class LabelSmoothingLoss(nn.Module):
    """
    With label smoothing,
    KL-divergence between q_{smoothed ground truth prob.}(w)
    and p_{prob. computed by model}(w) is minimized.
    """
    def __init__(self, label_smoothing, tgt_vocab_size, ignore_index=-100):
        assert 0.0 < label_smoothing <= 1.0
        self.ignore_index = ignore_index
        super(LabelSmoothingLoss, self).__init__()

        smoothing_value = label_smoothing / (tgt_vocab_size - 2)
        one_hot = torch.full((tgt_vocab_size,), smoothing_value)
        one_hot[self.ignore_index] = 0
        self.register_buffer('one_hot', one_hot.unsqueeze(0))

        self.confidence = 1.0 - label_smoothing

    def forward(self, output, target):
        """
        output (FloatTensor): batch_size x n_classes
        target (LongTensor): batch_size
        """
        model_prob = self.one_hot.repeat(target.size(0), 1)
        model_prob.scatter_(1, target.unsqueeze(1), self.confidence)
        model_prob.masked_fill_((target == self.ignore_index).unsqueeze(1), 0)

        return F.kl_div(output, model_prob, reduction='sum')

or is something else recommended?

obtained from:

references: