Question

我正在尝试使用＆＃39; tf.contrib.rnn.LayerNormBasicLSTMCell＆＃39;缩短训练LSTM细胞所需的时间。

在使用LayerNormBasicLSTMCell之前，我使用了＆＃tf.contrib.rnn.BasicLSTMCell＆＃39;，以下代码非常正常。

class Model():

    def __init__(self, args, infer=False):

        ''' Some definitions are here '''  

        cell_fn = tf.contrib.rnn.BasicLSTMCell

        def get_cell():
            return cell_fn(args.rnn_size)

        cell = get_cell()

        if (infer == False and args.keep_prob < 1):  # training mode
            cell = tf.contrib.rnn.DropoutWrapper(cell, output_keep_prob=args.keep_prob)
        self.cell = cell

        # Cell states : batch_size x (1, cell.state_size)
        zero_state = tf.split(tf.zeros([args.batch_size, cell.state_size]), axis=0, num_or_size_splits=args.batch_size)
        self.state_in = tf.identity(zero_state, name='state_in')
        self.state_out = tf.split(tf.zeros([args.batch_size, cell.state_size]), axis=0, num_or_size_splits=args.batch_size)
        self.output_states = tf.split(tf.zeros([args.batch_size, args.rnn_size]), axis=0, num_or_size_splits=args.batch_size)


        ''' Some definitions are here '''

        for b in range(args.batch_size):

            # current embedding sequence : (seq_length x rnn_size)
            current_emd_seq = embedding_seqs[b]

            for f in range(args.seq_length):

                # current embedding frame : (1 x rnn_size)
                current_emd_frame = tf.reshape(current_emd_seq[f], shape=(1, args.rnn_size))


                with tf.variable_scope("rnnlm") as scope:
                    if (b > 0 or f > 0):
                        scope.reuse_variables()

                # go through LSTM cell
                self.output_states[b], zero_state[b] = cell(current_emd_frame, zero_state[b])

但是，如果我刚从

更改了代码行

cell_fn = tf.contrib.rnn.BasicLSTMCell

到

cell_fn = tf.contrib.rnn.LayerNormBasicLSTMCell

发生以下错误......

Traceback (most recent call last):
  File "/home/dooseop/Projects/__IRLLSTM__/version20/kitti_train.py", 
line 192, in <module>
    main()
  File "/home/dooseop/Projects/__IRLLSTM__/version20/kitti_train.py", line 95, in main
    train(args)
  File "/home/dooseop/Projects/__IRLLSTM__/version20/kitti_train.py", line 112, in train
    model = Model(args)
  File "/home/dooseop/Projects/__IRLLSTM__/version20/kitti_model.py", line 193, in __init__
    self.output_states[b], zero_state[b] = cell(current_emd_frame, zero_state[b])
  File "/home/dooseop/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 1099, in __call__
    output, new_state = self._cell(inputs, state, scope=scope)
  File "/home/dooseop/anaconda3/lib/python3.6/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 232, in __call__
    return super(RNNCell, self).__call__(inputs, state)
  File "/home/dooseop/anaconda3/lib/python3.6/site-packages/tensorflow/python/layers/base.py", line 714, in __call__
    outputs = self.call(inputs, *args, **kwargs)
  File "/home/dooseop/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/rnn/python/ops/rnn_cell.py", line 1431, in 
call
    c, h = state
  File "/home/dooseop/anaconda3/lib/python3.6/site- 
packages/tensorflow/python/framework/ops.py", line 401, in __iter__
    "Tensor objects are not iterable when eager execution is not "
TypeError: Tensor objects are not iterable when eager execution is not 
enabled. To iterate over this tensor use tf.map_fn.

我也尝试过改变zero_state，self.state_out的大小，结果是一样的。

例如，更改

zero_state = tf.split(tf.zeros([args.batch_size, cell.state_size]), axis=0, num_or_size_splits=args.batch_size)

到

zero_state = tf.split(tf.zeros([args.batch_size, args.rnn_size, 2]), axis=0, num_or_size_splits=args.batch_size)

或

zero_state = tf.split(tf.zeros([args.batch_size, 2* args.rnn_size]), axis=0, num_or_size_splits=args.batch_size)

请告诉我使用方法的问题...谢谢！

如何在tensorflow中使用LayerNormBasicLSTMCell？

0 个答案: