Question

使用tf.get_collection()时，RNN单元格未显示。我错过了什么？

import tensorflow as tf
print(tf.__version__)

rnn_cell = tf.nn.rnn_cell.LSTMCell(16)
print(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))

other_var = tf.Variable(0)
print(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))

打印出来

0.12.0
[]
[<tensorflow.python.ops.variables.Variable object at 0x0000027961250B70>]

Windows 10，Python 3.5

Answer 1

您尚未在__call__上运行LSTMCell，这就是您没有看到变量的原因。试试这个（我假设batch_size=10和rnn_size=16）

import tensorflow as tf
print(tf.__version__)

rnn_cell = tf.nn.rnn_cell.LSTMCell(16)
a = tf.placeholder(tf.float32, [10, 16])
zero = rnn_cell.zero_state(10,tf.float32)
# The variables are created in the following __call__
b = rnn_cell(a, zero)
print(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))

other_var = tf.Variable(0)
print(tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES))

RNF Cell不存在于tf.get_collection中

1 个答案: