Question

我尝试在raw_rnn中为loop_fn使用自定义函数，但这有点奇怪

"raise TypeError("loop_fn must be a callable")"  # Exception thrown?

呼叫：

callable_loop_fn = loop_fn(
    time=time,
    previous_output=None,
    previous_state=None,
    previous_loop_state=None,
    _W=W, _b=b,
    _decoder_lengths=decoder_lengths,
    _pad_step_embedded=pad_step_embedded,
    _eos_step_embedded=eos_step_embedded,
    _encoder_final_state=encoder_final_state)

# using the functions for the attention decoder
decoder_outputs_ta, decoder_final_state, decoder_loop_state = tf.nn.raw_rnn(decoder_cell, callable_loop_fn)

定义：

def loop_fn(time, previous_output, previous_state, previous_loop_state, _W, _b, _decoder_lengths, _pad_step_embedded, _eos_step_embedded, _encoder_final_state):
if previous_state is None:
    assert previous_output is None and previous_state is None
    return loop_fn_initial(_decoder_lengths, _eos_step_embedded, _encoder_final_state)
else:
    return loop_fn_transition(time, previous_output, previous_state, previous_loop_state, _W, _b, _decoder_lengths, _pad_step_embedded)

有人知道这可能是什么吗？我认为我提供的功能是可调用的，还是我理解错了？

Answer 1

callable_loop_fn不是函数，因此它不可调用。

具体而言，callable_loop_fn是loop_fn()返回的值，而loop_fn_initial()则返回loop_fn_initial()的输出或loop_fn must be a callable的输出。显然，这两个函数都没有返回函数，因此抛出了异常def loop_fn(time, cell_output, cell_state, loop_state): ... return ( elements_finished, next_input, next_cell_state, emit_output, next_loop_state )。

根据TF API，你应该写：

tf.nn.raw_rnn

然后将其传递给raw_rnn(decoder_cell, loop_fn)：

loop_fn

请注意，您应该尊重Unexpected argument期望接收的参数的数量和顺序，否则您loop_fn会因函数add_dependencies(my_shared1 my_common_lib)而出错。因此，您的实现必须重新排列，只需要4个参数。

Tensorflow：tf.rnn.raw_rnn - fn_loop不可调用？

1 个答案: