I want to change the loss function in the ptb_word_lm.py example to tf.nn.nce_loss. Looking at the tf.nn.nce_loss implementation:
def nce_loss(weights, biases, inputs, labels, num_sampled, num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=False,
partition_strategy="mod",
name="nce_loss"):
I think
- the 3rd parameter (inputs) is the logits of language model,
- 4th parameter (labels) is the next word (self._targets) of language model,
- num_classes is the vocab_size
But I do not know what are the first two parameters, weights and biases. How could I adapt tf.nn.nce_loss to language model? Thanks.
@Aaron:
Thanks, I have tried the following:
loss = tf.reduce_mean(
tf.nn.nce_loss(softmax_w, softmax_b, logits, tf.reshape(self._targets, [-1,1]),
64, vocab_size))
According to the document at here:
weights: A Tensor of shape [num_classes, dim], or a list of Tensor 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.
num_sampled: An int. The number of classes to randomly sample per batch.
num_classes: An int. The number of possible classes.
So,
- weights is the softmax_w tensor, which has shape (hidden_size, vocab_size)
- biases is softmax_b, which has shape (vocab_size)
- inputs is logits, which has shape (batch_size*num_steps, vocab_size)
- labels is self._targets, which has shape (batch_size, num_steps), thus, we need to reshape it, tf.reshape(self._targets, [-1,1])
My PTBModel model looks like
class PTBModel(object):
def __init__(self, is_training, config):
self.batch_size = batch_size = config.batch_size
self.num_steps = num_steps = config.num_steps
size = config.hidden_size
vocab_size = config.vocab_size
self._input_data = tf.placeholder(tf.int32, [batch_size, num_steps])
self._targets = tf.placeholder(tf.int32, [batch_size, num_steps])
lstm_cell = rnn_cell.BasicLSTMCell(size, forget_bias=0.0)
if is_training and config.keep_prob < 1:
lstm_cell = rnn_cell.DropoutWrapper(lstm_cell, output_keep_prob=config.keep_prob)
cell = rnn_cell.MultiRNNCell([lstm_cell] * config.num_layers)
self._initial_state = cell.zero_state(batch_size, tf.float32)
with tf.device("/cpu:0"):
embedding = tf.get_variable("embedding", [vocab_size, size])
inputs = tf.nn.embedding_lookup(embedding, self._input_data)
if is_training and config.keep_prob < 1:
inputs = tf.nn.dropout(inputs, config.keep_prob)
outputs = []
states = []
state = self._initial_state
with tf.variable_scope("RNN"):
for time_step in range(num_steps):
if time_step > 0: tf.get_variable_scope().reuse_variables()
(cell_output, state) = cell(inputs[:, time_step, :], state)
outputs.append(cell_output)
states.append(state)
output = tf.reshape(tf.concat(1, outputs), [-1, size])
softmax_w = tf.get_variable("softmax_w", [size, vocab_size])
softmax_b = tf.get_variable("softmax_b", [vocab_size])
logits = tf.matmul(output, softmax_w) + softmax_b
'''
#minimize the average negative log probability using sequence_loss_by_example
loss = seq2seq.sequence_loss_by_example([logits],
[tf.reshape(self._targets, [-1])],
[tf.ones([batch_size * num_steps])],
vocab_size)
loss = tf.reduce_mean(
tf.nn.nce_loss(nce_weights, nce_biases, embed, train_labels,
num_sampled, vocabulary_size))
weights: A Tensor of shape [num_classes, dim], or a list of Tensor 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.
num_sampled: An int. The number of classes to randomly sample per batch.
num_classes: An int. The number of possible classes.
'''
loss = tf.reduce_mean(
tf.nn.nce_loss(softmax_w, softmax_b, logits, tf.reshape(self._targets, [-1,1]),
64, vocab_size))
self._cost = cost = tf.reduce_sum(loss) / batch_size
self._final_state = states[-1]
if not is_training:
return
self._lr = tf.Variable(0.0, trainable=False)
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(cost, tvars),
config.max_grad_norm)
optimizer = tf.train.GradientDescentOptimizer(self.lr)
self._train_op = optimizer.apply_gradients(zip(grads, tvars))
However, I got an error
Epoch: 1 Learning rate: 1.000
W tensorflow/core/common_runtime/executor.cc:1102] 0x528c980 Compute status: Invalid argument: Index 9971 at offset 0 in Tindices is out of range
[[Node: model/nce_loss/embedding_lookup = Gather[Tindices=DT_INT64, Tparams=DT_FLOAT, validate_indices=true, _device="/job:localhost/replica:0/task:0/cpu:0"](model/softmax_w/read, model/nce_loss/concat)]]
W tensorflow/core/common_runtime/executor.cc:1102] 0x528c980 Compute status: Invalid argument: Index 9971 at offset 0 in Tindices is out of range
[[Node: model/nce_loss/embedding_lookup = Gather[Tindices=DT_INT64, Tparams=DT_FLOAT, validate_indices=true, _device="/job:localhost/replica:0/task:0/cpu:0"](model/softmax_w/read, model/nce_loss/concat)]]
[[Node: _send_model/RNN/concat_19_0 = _Send[T=DT_FLOAT, client_terminated=true, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/cpu:0", send_device_incarnation=1438650956868917036, tensor_name="model/RNN/concat_19:0", _device="/job:localhost/replica:0/task:0/cpu:0"](model/RNN/concat_19)]]
Traceback (most recent call last):
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 235, in <module>
tf.app.run()
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/default/_app.py", line 30, in run
sys.exit(main(sys.argv))
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 225, in main
verbose=True)
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 189, in run_epoch
m.initial_state: state})
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 315, in run
return self._run(None, fetches, feed_dict)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 511, in _run
feed_dict_string)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 564, in _do_run
target_list)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 586, in _do_call
e.code)
tensorflow.python.framework.errors.InvalidArgumentError: Index 9971 at offset 0 in Tindices is out of range
[[Node: model/nce_loss/embedding_lookup = Gather[Tindices=DT_INT64, Tparams=DT_FLOAT, validate_indices=true, _device="/job:localhost/replica:0/task:0/cpu:0"](model/softmax_w/read, model/nce_loss/concat)]]
Caused by op u'model/nce_loss/embedding_lookup', defined at:
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 235, in <module>
tf.app.run()
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/platform/default/_app.py", line 30, in run
sys.exit(main(sys.argv))
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 214, in main
m = PTBModel(is_training=True, config=config)
File "/home/user/works/workspace/python/ptb_word_lm/ptb_word_lm.py", line 122, in __init__
64, vocab_size))
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/nn.py", line 798, in nce_loss
name=name)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/nn.py", line 660, in _compute_sampled_logits
weights, all_ids, partition_strategy=partition_strategy)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/embedding_ops.py", line 86, in embedding_lookup
validate_indices=validate_indices)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_array_ops.py", line 447, in gather
validate_indices=validate_indices, name=name)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/op_def_library.py", line 655, in apply_op
op_def=op_def)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 2040, in create_op
original_op=self._default_original_op, op_def=op_def)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1087, in __init__
self._traceback = _extract_stack()
Did I miss anything here? Thanks again.
