losses.py 文件源码

def calculate_loss(self, predictions, labels, **unused_params):
    false_positive_punishment = FLAGS.false_positive_punishment
    false_negative_punishment = FLAGS.false_negative_punishment
    with tf.name_scope("loss_xent_recall"):
      epsilon = 10e-6
      if FLAGS.label_smoothing:
        float_labels = smoothing(labels)
      else:
        float_labels = tf.cast(labels, tf.float32)
      cross_entropy_loss = false_negative_punishment * float_labels * tf.log(predictions + epsilon) \
          + false_positive_punishment * ( 1 - float_labels) * tf.log(1 - predictions + epsilon)
      cross_entropy_loss = tf.negative(cross_entropy_loss)
      return tf.reduce_mean(tf.reduce_sum(cross_entropy_loss, 1))