请求tensorflow = float64_ref，actual = float64

Question

我只是想让一个简单的RNNCell工作。这个简单的代码：

with tf.Session() as sess:
    x = tf.Variable(np.ones((2, 3)))
    tf.initialize_all_variables().run()
    out, state = BasicRNNCell(4)(x, x)

引发以下错误：

Traceback (most recent call last):
  File "scrap.py", line 38, in <module>
    g, _ = tf.nn.rnn_cell.BasicRNNCell(2)(x, x)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell.py", line 199, in __call__
    output = self._activation(_linear([inputs, state], self._num_units, True))
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell.py", line 903, in _linear
    "Matrix", [total_arg_size, output_size], dtype=dtype)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 1022, in get_variable
    custom_getter=custom_getter)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 849, in get_variable
    custom_getter=custom_getter)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 345, in get_variable
    validate_shape=validate_shape)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 330, in _true_getter
    caching_device=caching_device, validate_shape=validate_shape)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 676, in _get_single_variable
    validate_shape=validate_shape)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variables.py", line 215, in __init__
    dtype=dtype)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variables.py", line 288, in _init_from_args
    initial_value(), name="initial_value", dtype=dtype)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 666, in <lambda>
    shape.as_list(), dtype=dtype, partition_info=partition_info)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/init_ops.py", line 280, in _initializer
    dtype, seed=seed)
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/ops/random_ops.py", line 232, in random_uniform
    minval = ops.convert_to_tensor(minval, dtype=dtype, name="min")
  File "/Users/ethan/env/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 671, in convert_to_tensor
    dtype.name, ret.dtype.name))
RuntimeError: min: Conversion function <function _constant_tensor_conversion_function at 0x112053c08> for type <type 'object'> returned incompatible dtype: requested = float64_ref, actual = float64

这是pip show tensorflow的输出：

元数据 - 版本：2.0名称：tensorflow版本：0.11.0rc1摘要：TensorFlow帮助张量流程主页：http://tensorflow.org/作者：Google Inc.作者电子邮件：[email protected]安装程序：pip许可证：Apache 2.0位置：/ Users /ethan/env/lib/python2.7/site-packages需要：mock，protobuf，numpy，wheel，six分类器：开发状态:: 4 - Beta预期受众::开发人员目标受众::教育目标受众::科学/研究许可证:: OSI批准:: Apache软件许可证编程语言:: Python :: 2.7主题::科学/工程::数学主题::软件开发::库:: Python模块主题::软件开发::库输入 - points：[console_scripts] tensorboard = tensorflow.tensorboard.tensorboard：main

谢谢！

Answer 1

在TensorFlow中，类Tensor的dtype为float64_ref，类Variable的dtype为float64。并且BasicRNNCell需要除变量之外的Tensor作为输入。在您的代码中，x = tf.Variable(...)应更改为x = tf.convert_to_tensor(...)

Answer 2

我认为这是Tensorflow中的一个错误，我邀请你加入open an issue on GitHub。

要解决此问题，您可以使用tf.identity创建float64_ref而不是float64 x并将此值作为inputs参数传递。

import tensorflow as tf
import numpy as np

with tf.Session() as sess:
    x = tf.Variable(np.ones((2, 3)))
    sess.run(tf.initialize_all_variables())
    out, state = tf.nn.rnn_cell.BasicRNNCell(4)(tf.identity(x), x)