Keras导入错误Tensorflow后端

Question

    ---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
/Users/CJL/anaconda3/lib/python3.5/site-packages/keras/backend/tensorflow_backend.py in <module>()
      4 try:
----> 5     from tensorflow.python.ops import ctc_ops as ctc
      6 except ImportError:

ImportError: cannot import name 'ctc_ops'

During handling of the above exception, another exception occurred:

ImportError                               Traceback (most recent call last)
<ipython-input-10-c74e2bd4ca71> in <module>()
----> 1 import keras

/Users/CJL/anaconda3/lib/python3.5/site-packages/keras/__init__.py in <module>()
      1 from __future__ import absolute_import
----> 2 from . import backend
      3 from . import datasets
      4 from . import engine
      5 from . import layers

/Users/CJL/anaconda3/lib/python3.5/site-packages/keras/backend/__init__.py in <module>()
     67 elif _BACKEND == 'tensorflow':
     68     sys.stderr.write('Using TensorFlow backend.\n')
---> 69     from .tensorflow_backend import *
     70 else:
     71     raise Exception('Unknown backend: ' + str(_BACKEND))

/Users/CJL/anaconda3/lib/python3.5/site-packages/keras/backend/tensorflow_backend.py in <module>()
      5     from tensorflow.python.ops import ctc_ops as ctc
      6 except ImportError:
----> 7     import tensorflow.contrib.ctc as ctc
      8 
      9 import numpy as np

ImportError: No module named 'tensorflow.contrib.ctc'

似乎是Keras的问题，而不是Tensorflow的问题，根据这篇文章在这里输入链接描述， 我想知道是否已解决？

Answer 1

这是Keras和张量流之间的问题。您应该将此修复程序更改为使用类似

if hasattr(tf.contrib, 'ctc'):
  ctc = tf.contrib.ctc  # old version
else:
  ctc = tf.nn  # New official version.

目前的“官方”的方式来访问CTC操作是tf.nn，例如，tf.nn.ctc_loss。