ValueError: Registered op_def for CudnnRNN (name: "CudnnRNN"

问题

File "D:\Software\anaconda3-5.0\envs\ame\lib\site-packages\tensorflow_core\python\framework\op_def_registry.py", line 38, in register_op_list
    % (op_def.name, _registered_ops[op_def.name], op_def))
ValueError: Registered op_def for CudnnRNN (name: "CudnnRNN"
input_arg {
  name: "input"
  type_attr: "T"
}
input_arg {
  name: "input_h"
  type_attr: "T"
}
input_arg {
  name: "input_c"
  type_attr: "T"
}
input_arg {
  name: "params"
  type_attr: "T"
}
output_arg {
  name: "output"
  type_attr: "T"
}
output_arg {
  name: "output_h"
  type_attr: "T"
}
output_arg {
  name: "output_c"
  type_attr: "T"
}
output_arg {
  name: "reserve_space"
  type_attr: "T"
}
attr {
  name: "T"
  type: "type"
  allowed_values {
    list {
      type: DT_HALF
      type: DT_FLOAT
      type: DT_DOUBLE
    }
  }
}
attr {
  name: "rnn_mode"
  type: "string"
  default_value {
    s: "lstm"
  }
  allowed_values {
    list {
      s: "rnn_relu"
      s: "rnn_tanh"
      s: "lstm"
      s: "gru"
    }
  }
}
attr {
  name: "input_mode"
  type: "string"
  default_value {
    s: "linear_input"
  }
  allowed_values {
    list {
      s: "linear_input"
      s: "skip_input"
      s: "auto_select"
    }
  }
}
attr {
  name: "direction"
  type: "string"
  default_value {
    s: "unidirectional"
  }
  allowed_values {
    list {
      s: "unidirectional"
      s: "bidirectional"
    }
  }
}
attr {
  name: "dropout"
  type: "float"
  default_value {
    f: 0.0
  }
}
attr {
  name: "seed"
  type: "int"
  default_value {
    i: 0
  }
}
attr {
  name: "seed2"
  type: "int"
  default_value {
    i: 0
  }
}
attr {
  name: "is_training"
  type: "bool"
  default_value {
    b: true
  }
}
is_stateful: true
) not equal to op_def to register (name: "CudnnRNN"
input_arg {
  name: "input"
  type_attr: "T"
}
input_arg {
  name: "input_h"
  type_attr: "T"
}
input_arg {
  name: "input_c"
  type_attr: "T"
}
input_arg {
  name: "params"
  type_attr: "T"
}
output_arg {
  name: "output"
  type_attr: "T"
}
output_arg {
  name: "output_h"
  type_attr: "T"
}
output_arg {
  name: "output_c"
  type_attr: "T"
}
output_arg {
  name: "reserve_space"
  type_attr: "T"
}
attr {
  name: "T"
  type: "type"
  allowed_values {
    list {
      type: DT_FLOAT
    }
  }
}
attr {
  name: "rnn_mode"
  type: "string"
  default_value {
    s: "lstm"
  }
  allowed_values {
    list {
      s: "rnn_relu"
      s: "rnn_tanh"
      s: "lstm"
      s: "gru"
    }
  }
}
attr {
  name: "input_mode"
  type: "string"
  default_value {
    s: "auto_select"
  }
  allowed_values {
    list {
      s: "linear_input"
      s: "skip_input"
      s: "auto_select"
    }
  }
}
attr {
  name: "direction"
  type: "string"
  default_value {
    s: "unidirectional"
  }
  allowed_values {
    list {
      s: "unidirectional"
      s: "bidirectional"
    }
  }
}
attr {
  name: "dropout"
  type: "float"
  default_value {
    f: 0.0
  }
}
attr {
  name: "seed"
  type: "int"
  default_value {
    i: 0
  }
}
attr {
  name: "seed2"
  type: "int"
  default_value {
    i: 0
  }
}
attr {
  name: "is_training"
  type: "bool"
  default_value {
    b: true
  }
}
is_stateful: true
)

解决方案

在op_def_registery中加入如下

from tensorflow.contrib.cudnn_rnn.ops import gen_cudnn_rnn_ops