Have a question? Connect with the community at the TensorFlow Forum Visit Forum

tfnlp.losses.weighted_sparse_categorical_crossentropy_loss

Calculate a per-batch sparse categorical crossentropy loss.

This loss function assumes that the predictions are post-softmax. Args: labels: The labels to evaluate against. Should be a set of integer indices ranging from 0 to (vocab_size-1). predictions: The network predictions. Should have softmax already applied. weights: An optional weight array of the same shape as the 'labels' array. If None, all examples will be used. from_logits: Whether the input predictions are logits.

A loss scalar.

RuntimeError if the passed tensors do not have the same rank.