Tensorflow：噪声对比估计语言模型

我想将ptb_word_lm.py示例中的损失函数更改为 tf.nn.nce_loss . 查看 tf.nn.nce_loss 实现：

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"):

我认为

• 第3个参数（输入）是语言模型的logits，

• 第四个参数（标签）是语言模型的下一个单词（self._targets），

• num_classes是vocab_size

但我不知道前两个参数，权重和偏差是什么 . 我怎样才能使 tf.nn.nce_loss 适应语言模型？谢谢 .
＃＃＃＃＃＃＃＃更新
@Aaron：

谢谢，我尝试过以下方法：

loss = tf.reduce_mean(
        tf.nn.nce_loss(softmax_w, softmax_b, logits, tf.reshape(self._targets, [-1,1]),
                                     64, vocab_size))

根据here的文件：

• 权重：形状张量[num_classes，dim]或Tensor对象列表，其沿着维度0的连接具有形状[num_classes，dim] . （可能是分区的）类嵌入 .

• 偏差：形状张量[num_classes] . class 偏见 .

• 输入：形状张量[batch_size，dim] . 输入网络的前向激活 .

• 标签：类型为int64且形状为[batch_size，num_true]的张量 . 目标类 .

• num_sampled：一个int . 每批随机抽样的类数 .

• num_classes：一个int . 可能的类数 .

所以，

• 权重是softmax_w张量，它有形状（hidden_size，vocab_size）

• 偏差是softmax_b，它有形状（vocab_size）

• 输入是logits，具有形状（batch_size * num_steps，vocab_size）

• 标签是self._targets，它有shape（batch_size，num_steps），因此，我们需要重塑它，tf.reshape（self._targets，[ - 1,1]）

我的PTBModel模型看起来像

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))

但是，我收到了一个错误

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()

我在这里错过了吗？再次感谢 .