Swap the input and label arguments in nce_loss

Change: 141244045

tensorflow/examples/tutorials/word2vec/word2vec_basic.py

@@ -160,8 +160,12 @@ with graph.as_default():
   # tf.nce_loss automatically draws a new sample of the negative labels each
   # time we evaluate the loss.
   loss = tf.reduce_mean(
-      tf.nn.nce_loss(nce_weights, nce_biases, embed, train_labels,
+      tf.nn.nce_loss(weights=nce_weights,
-                     num_sampled, vocabulary_size))
+                     biases=nce_biases,
+                     labels=train_labels,
+                     inputs=embed,
+                     num_sampled=num_sampled,
+                     num_classes=vocabulary_size))
 
   # Construct the SGD optimizer using a learning rate of 1.0.
   optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(loss)

tensorflow/python/ops/nn_impl.py

@@ -809,8 +809,8 @@ def _sum_rows(x):
 
 def _compute_sampled_logits(weights,
                             biases,
-                            inputs,
                             labels,
+                            inputs,
                             num_sampled,
                             num_classes,
                             num_true=1,
@@ -834,11 +834,11 @@ def _compute_sampled_logits(weights,
         objects whose concatenation along dimension 0 has shape
         `[num_classes, dim]`.  The (possibly-partitioned) class embeddings.
     biases: A `Tensor` of shape `[num_classes]`.  The class biases.
-    inputs: A `Tensor` of shape `[batch_size, dim]`.  The forward
-        activations of the input network.
     labels: A `Tensor` of type `int64` and shape `[batch_size,
         num_true]`. The target classes.  Note that this format differs from
         the `labels` argument of `nn.softmax_cross_entropy_with_logits`.
+    inputs: A `Tensor` of shape `[batch_size, dim]`.  The forward
+        activations of the input network.
     num_sampled: An `int`.  The number of classes to randomly sample per batch.
     num_classes: An `int`. The number of possible classes.
     num_true: An `int`.  The number of target classes per training example.
@@ -975,8 +975,8 @@ def _compute_sampled_logits(weights,
 
 def nce_loss(weights,
              biases,
-             inputs,
              labels,
+             inputs,
              num_sampled,
              num_classes,
              num_true=1,
@@ -1012,10 +1012,10 @@ def nce_loss(weights,
         objects whose concatenation along dimension 0 has shape
         [num_classes, dim].  The (possibly-partitioned) class embeddings.
     biases: A `Tensor` of shape `[num_classes]`.  The class biases.
-    inputs: A `Tensor` of shape `[batch_size, dim]`.  The forward
-        activations of the input network.
     labels: A `Tensor` of type `int64` and shape `[batch_size,
         num_true]`. The target classes.
+    inputs: A `Tensor` of shape `[batch_size, dim]`.  The forward
+        activations of the input network.
     num_sampled: An `int`.  The number of classes to randomly sample per batch.
     num_classes: An `int`. The number of possible classes.
     num_true: An `int`.  The number of target classes per training example.
@@ -1038,12 +1038,12 @@ def nce_loss(weights,
     A `batch_size` 1-D tensor of per-example NCE losses.
   """
   logits, labels = _compute_sampled_logits(
-      weights,
+      weights=weights,
-      biases,
+      biases=biases,
-      inputs,
+      labels=labels,
-      labels,
+      inputs=inputs,
-      num_sampled,
+      num_sampled=num_sampled,
-      num_classes,
+      num_classes=num_classes,
       num_true=num_true,
       sampled_values=sampled_values,
       subtract_log_q=True,
@@ -1114,12 +1114,12 @@ def sampled_softmax_loss(weights,
 
   """
   logits, labels = _compute_sampled_logits(
-      weights,
+      weights=weights,
-      biases,
+      biases=biases,
-      inputs,
+      labels=labels,
-      labels,
+      inputs=inputs,
-      num_sampled,
+      num_sampled=num_sampled,
-      num_classes,
+      num_classes=num_classes,
       num_true=num_true,
       sampled_values=sampled_values,
       subtract_log_q=True,

tensorflow/python/ops/nn_test.py

@@ -452,8 +452,8 @@ class ComputeSampledLogitsTest(tf.test.TestCase):
     pred_logits_tf, pred_labels_tf = _compute_sampled_logits(
         weights_tf,
         biases_tf,
-        hidden_acts_tf,
         labels_tf,
+        hidden_acts_tf,
         num_sampled,
         num_classes,
         num_true,
@@ -672,8 +672,8 @@ class ComputeSampledLogitsTest(tf.test.TestCase):
       nce_loss_tf = tf.nn.nce_loss(
           weights_tf,
           biases_tf,
-          inputs_tf,
           labels_tf,
+          inputs_tf,
           num_sampled=1,
           num_classes=self._num_classes,
           num_true=1,
@@ -685,8 +685,8 @@ class ComputeSampledLogitsTest(tf.test.TestCase):
       nce_loss_tf = tf.nn.nce_loss(
           [tf.constant(shard) for shard in sharded_weights],
           biases_tf,
-          inputs_tf,
           labels_tf,
+          inputs_tf,
           num_sampled=1,
           num_classes=self._num_classes,
           num_true=1,