python - lstm 维度与 tensorflow 不匹配

Question

我构建了一个 LSTM 网络，输入的维度为 100*100*83 (batch_size=100, steps = 100, char_vector = 83)。我构建了两个 LSTM 层，其中有 512 个隐藏单元。

# coding: utf-8

from __future__ import print_function
import tensorflow as tf
import numpy as np
import time


class CharRNN:
    def __init__(self, num_classes, batch_size=64, num_steps=50, lstm_size=128, num_layers =2,\
                 learning_rate = 0.001, grad_clip=5, keep_prob=0.001,sampling= False):
        # True for SGD
        if sampling == True:
            self.batch_size, self.num_steps = 1,1
        else:
            self.batch_size, self.num_steps = batch_size, num_steps

        tf.reset_default_graph()

        self.inputs, self.targets, self.keep_prob = self.build_inputs(self.batch_size,self.num_steps)
        self.keep_prob = keep_prob

        self.cell, self.initial_state = self.build_lstm(lstm_size,num_layers,self.batch_size,self.keep_prob)

       # print(self.cell.state_size)
        x_one_hot = tf.one_hot(self.inputs, num_classes)
        print("cell state size: ",self.cell.state_size)
        print("cell initial  state: ",self.initial_state)
        print("this is inputs", self.inputs)

        print("x_one_hot: ",x_one_hot)
        outputs, state = tf.nn.dynamic_rnn(self.cell, x_one_hot, initial_state= self.initial_state)



    def build_inputs(self, num_seqs, num_steps):
        inputs = tf.placeholder(tf.int32, shape=(num_seqs, num_steps), name = "inputs")
        targets = tf.placeholder(tf.int32, shape= (num_seqs, num_steps), name="targets")
        print('inputs shape: ',inputs.shape)
        keep_prob = tf.placeholder(tf.float32, name="keep_prob")
        return inputs, targets, keep_prob

    def build_lstm(self, lstm_size, num_layers, batch_size, keep_prob):

        # construct lstm cell
        lstm = tf.nn.rnn_cell.BasicLSTMCell(lstm_size)
        # add dropout
        drop = tf.nn.rnn_cell.DropoutWrapper(lstm, output_keep_prob= keep_prob)

        # stack multiple rnn cells
        cell = tf.nn.rnn_cell.MultiRNNCell([drop for _ in range(num_layers)])
        initial_state = cell.zero_state(batch_size, tf.float32)

        return cell, initial_state

if __name__ == '__main__':
    len_vocab = 83
    batch_size = 100
    num_steps = 100
    lstm_size = 512
    num_layers = 2
    learning_rate = 0.001
    keep_prob = 0.5

    epochs = 20
    save_every_n = 200

    print("h1")
    model = CharRNN(len_vocab, batch_size = batch_size, num_steps=num_steps, lstm_size = lstm_size,num_layers=num_layers\
                    ,learning_rate=learning_rate,sampling= False,keep_prob = keep_prob

我在 tf.nn.dynamic_rnn 处收到尺寸不匹配错误。错误信息是这样的:

    inputs shape:  (100, 100)
cell state size:  (LSTMStateTuple(c=512, h=512), LSTMStateTuple(c=512, h=512))
cell initial  state:  (LSTMStateTuple(c=<tf.Tensor 'MultiRNNCellZeroState/DropoutWrapperZeroState/BasicLSTMCellZeroState/zeros:0' shape=(100, 512) dtype=float32>, h=<tf.Tensor 'MultiRNNCellZeroState/DropoutWrapperZeroState/BasicLSTMCellZeroState/zeros_1:0' shape=(100, 512) dtype=float32>), LSTMStateTuple(c=<tf.Tensor 'MultiRNNCellZeroState/DropoutWrapperZeroState_1/BasicLSTMCellZeroState/zeros:0' shape=(100, 512) dtype=float32>, h=<tf.Tensor 'MultiRNNCellZeroState/DropoutWrapperZeroState_1/BasicLSTMCellZeroState/zeros_1:0' shape=(100, 512) dtype=float32>))
this is inputs Tensor("inputs:0", shape=(100, 100), dtype=int32)
x_one_hot:  Tensor("one_hot:0", shape=(100, 100, 83), dtype=float32)
Traceback (most recent call last):
  File "./seq2_minimal.py", line 70, in <module>
    ,learning_rate=learning_rate,sampling= False,keep_prob = keep_prob)
  File "./seq2_minimal.py", line 32, in __init__
    outputs, state = tf.nn.dynamic_rnn(self.cell, x_one_hot, initial_state= self.initial_state)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn.py", line 614, in dynamic_rnn
    dtype=dtype)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn.py", line 777, in _dynamic_rnn_loop
    swap_memory=swap_memory)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/control_flow_ops.py", line 2816, in while_loop
    result = loop_context.BuildLoop(cond, body, loop_vars, shape_invariants)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/control_flow_ops.py", line 2640, in BuildLoop
    pred, body, original_loop_vars, loop_vars, shape_invariants)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/control_flow_ops.py", line 2590, in _BuildLoop
    body_result = body(*packed_vars_for_body)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn.py", line 762, in _time_step
    (output, new_state) = call_cell()
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn.py", line 748, in <lambda>
    call_cell = lambda: cell(input_t, state)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 183, in __call__
    return super(RNNCell, self).__call__(inputs, state)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/layers/base.py", line 575, in __call__
    outputs = self.call(inputs, *args, **kwargs)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 1066, in call
    cur_inp, new_state = cell(cur_inp, cur_state)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 891, in __call__
    output, new_state = self._cell(inputs, state, scope)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 183, in __call__
    return super(RNNCell, self).__call__(inputs, state)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/layers/base.py", line 575, in __call__
    outputs = self.call(inputs, *args, **kwargs)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 441, in call
    value=self._linear([inputs, h]), num_or_size_splits=4, axis=1)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/rnn_cell_impl.py", line 1189, in __call__
    res = math_ops.matmul(array_ops.concat(args, 1), self._weights)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/math_ops.py", line 1891, in matmul
    a, b, transpose_a=transpose_a, transpose_b=transpose_b, name=name)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 2437, in _mat_mul
    name=name)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
    op_def=op_def)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2958, in create_op
    set_shapes_for_outputs(ret)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2209, in set_shapes_for_outputs
    shapes = shape_func(op)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2159, in call_with_requiring
    return call_cpp_shape_fn(op, require_shape_fn=True)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 627, in call_cpp_shape_fn
    require_shape_fn)
  File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 691, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Dimensions must be equal, but are 1024 and 595 for 'rnn/while/rnn/multi_rnn_cell/cell_0/cell_0/basic_lstm_cell/MatMul_1' (op: 'MatMul') with input shapes: [100,1024], [595,2048].

我搜索并发现tensorflow的lstm单元应该自动调整其输入大小。但错误消息是这样说的。

显示

input size is [100, 1024] and lstm is [595, 2048].

首先谢谢。

Answer 1

cell = tf.nn.rnn_cell.MultiRNNCell([drop for _ in range(num_layers)])

致

cell = tf.nn.rnn_cell.MultiRNNCell([drop])

因为您给定的输入张量和生成张量不同。