5ec08ed0e16d87c029bc5d9de46b95800d0bc470,texar/losses/mle_losses.py,,smoothing_cross_entropy,#Any#Any#Any#Any#Any#,169

Before Change

    // We subtract it just for readability, makes no difference on learning.
    normalizing = -(
        confidence * tf.log(confidence) + tf.to_float(vocab_size - 1) *
        low_confidence * tf.log(low_confidence + 1e-20))

    if gaussian and confidence > 0.0:
      labels = tf.cast(labels, tf.float32)
      normal_dist = tf.distributions.Normal(loc=labels, scale=confidence)
      // Locations to evaluate the probability distributions.
      soft_targets = normal_dist.prob(
          tf.cast(tf.range(vocab_size), tf.float32)[:, None, None, None, None])
      // Reordering soft_targets from [vocab_size, batch_size, ?, ?, ?] to match
      // logits: [batch_size, ?, ?, ?, vocab_size]
      soft_targets = tf.transpose(soft_targets, perm=[1, 2, 3, 4, 0])
    else:
      soft_targets = tf.one_hot(
          tf.cast(labels, tf.int32),
          depth=vocab_size,
          on_value=confidence,
          off_value=low_confidence)
    xentropy = tf.nn.softmax_cross_entropy_with_logits(
        logits=logits, labels=soft_targets)
    return xentropy - normalizing

After Change

        cross_entropy_fn = tf.nn.softmax_cross_entropy_with_logits_v2
    else:
        cross_entropy_fn = tf.nn.softmax_cross_entropy_with_logits
    return cross_entropy_fn(
        logits=logits, labels=soft_targets)

Instances