tf.nn.nce_loss
tf.nn.nce_loss
tf.nn.nce_loss
nce_loss( weights, biases, labels, inputs, num_sampled, num_classes, num_true=1, sampled_values=None, remove_accidental_hits=False, partition_strategy='mod', name='nce_loss' )
Defined in tensorflow/python/ops/nn_impl.py
.
See the guide: Neural Network > Candidate Sampling
Computes and returns the noise-contrastive estimation training loss.
See Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. Also see our Candidate Sampling Algorithms Reference
A common use case is to use this method for training, and calculate the full sigmoid loss for evaluation or inference. In this case, you must set partition_strategy="div"
for the two losses to be consistent, as in the following example:
if mode == "train": loss = tf.nn.nce_loss( weights=weights, biases=biases, labels=labels, inputs=inputs, ..., partition_strategy="div") elif mode == "eval": logits = tf.matmul(inputs, tf.transpose(weights)) logits = tf.nn.bias_add(logits, biases) labels_one_hot = tf.one_hot(labels, n_classes) loss = tf.nn.sigmoid_cross_entropy_with_logits( labels=labels_one_hot, logits=logits) loss = tf.reduce_sum(loss, axis=1)
Note: By default this uses a log-uniform (Zipfian) distribution for sampling, so your labels must be sorted in order of decreasing frequency to achieve good results. For more details, see tf.nn.log_uniform_candidate_sampler
.
Note: In the case wherenum_true
> 1, we assign to each target class the target probability 1 /num_true
so that the target probabilities sum to 1 per-example.
Note: It would be useful to allow a variable number of target classes per example. We hope to provide this functionality in a future release. For now, if you have a variable number of target classes, you can pad them out to a constant number by either repeating them or by padding with an otherwise unused class.
Args:
-
weights
: ATensor
of shape[num_classes, dim]
, or a list ofTensor
objects whose concatenation along dimension 0 has shape [num_classes, dim]. The (possibly-partitioned) class embeddings. -
biases
: ATensor
of shape[num_classes]
. The class biases. -
labels
: ATensor
of typeint64
and shape[batch_size, num_true]
. The target classes. -
inputs
: ATensor
of shape[batch_size, dim]
. The forward activations of the input network. -
num_sampled
: Anint
. The number of classes to randomly sample per batch. -
num_classes
: Anint
. The number of possible classes. -
num_true
: Anint
. The number of target classes per training example. -
sampled_values
: a tuple of (sampled_candidates
,true_expected_count
,sampled_expected_count
) returned by a*_candidate_sampler
function. (if None, we default tolog_uniform_candidate_sampler
) -
remove_accidental_hits
: Abool
. Whether to remove "accidental hits" where a sampled class equals one of the target classes. If set toTrue
, this is a "Sampled Logistic" loss instead of NCE, and we are learning to generate log-odds instead of log probabilities. See our [Candidate Sampling Algorithms Reference] (https://www.tensorflow.org/extras/candidate_sampling.pdf). Default is False. -
partition_strategy
: A string specifying the partitioning strategy, relevant iflen(weights) > 1
. Currently"div"
and"mod"
are supported. Default is"mod"
. Seetf.nn.embedding_lookup
for more details. -
name
: A name for the operation (optional).
Returns:
A batch_size
1-D tensor of per-example NCE losses.
© 2017 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.
https://www.tensorflow.org/api_docs/python/tf/nn/nce_loss