我想用kest中的lstm实现卷积递归神经网络。我是机器学习的初学者，因此我很难理解一切。这是我的代码：
def initialize_model():
'Function to initialize the Convolutional Neural Network'
seed=123
# define CNN model
model = Sequential()
model.add(TimeDistributed(Conv2D(input_shape=(None,1,2048,1),kernel_size=(256,1),strides=(32,1),filters=16
,activation=LeakyReLU(0.01),data_format='channels_first')))
#cnn.add(MaxPooling2D(...))
model.add(TimeDistributed(Flatten()))
# define LSTM model
#model = Sequential()
#model.add(TimeDistribu(cnn))
model.add(LSTM(1024,recurrent_activation=LeakyReLU(0.01),recurrent_initializer=glorot_normal(seed)
,recurrent_regularizer=regularizers.l2(0.005)))
model.add(Dense(2048))
return model
def compile_model(model,learning_rate,loss_function):
#optimizer
optimizer = Adam(lr=learning_rate,beta_1=0.9,beta_2=0.999,epsilon=1e-8)
#Compile the model
model.compile(optimizer=optimizer,loss=loss_function,metrics=["accuracy"])
return model
# import and split data
data = get_data_from_files(10)
print(data.shape)
X_train, X_test, y_train, y_test = train_test_split(data[0],data[1], test_size=0.2, random_state=123)
print(X_train.shape) -> 160000,2048
print(y_train.shape) -> 160000,2048
print(X_test.shape) -> 40000,2048
print(y_test.shape) -> 40000,2048
X_tr = np.expand_dims(np.expand_dims(X_train,axis=2),axis=3)
X_te = np.expand_dims(np.expand_dims(X_test,axis=2),axis=3)
#X_tr shape -->(160000, 2048, 1, 1)
#X_te shape -->(40000, 2048, 1, 1)
model = initialize_model()
model =compile_model(model,0.001,"cosine_proximity")
hist = model.fit(X_tr,y_train
,validation_data=[X_te,y_test]
,epochs=5,batch_size=1000)
它返回以下错误：
ValueError Traceback (most recent call last)
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py in merge_with(self, other)
666 try:
--> 667 self.assert_same_rank(other)
668 new_dims = []
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py in assert_same_rank(self, other)
711 raise ValueError("Shapes %s and %s must have the same rank" % (self,
--> 712 other))
713
ValueError: Shapes (256, 1, 2048, 16) and (?, ?, ?) must have the same rank
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py in with_rank(self, rank)
741 try:
--> 742 return self.merge_with(unknown_shape(ndims=rank))
743 except ValueError:
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py in merge_with(self, other)
672 except ValueError:
--> 673 raise ValueError("Shapes %s and %s are not compatible" % (self, other))
674
ValueError: Shapes (256, 1, 2048, 16) and (?, ?, ?) are not compatible
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
<ipython-input-36-1633649265a0> in <module>()
6 hist = model.fit(X_tr,y_train
7 ,validation_data=[X_te,y_test]
----> 8 ,epochs=5,batch_size=1000)
9
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
953 sample_weight=sample_weight,
954 class_weight=class_weight,
--> 955 batch_size=batch_size)
956 # Prepare validation data.
957 do_validation = False
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/engine/training.py in _standardize_user_data(self, x, y, sample_weight, class_weight, check_array_lengths, batch_size)
674 # to match the value shapes.
675 if not self.inputs:
--> 676 self._set_inputs(x)
677
678 if y is not None:
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/engine/training.py in _set_inputs(self, inputs, outputs, training)
574 assert len(inputs) == 1
575 inputs = inputs[0]
--> 576 self.build(input_shape=(None,) + inputs.shape[1:])
577 return
578
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/engine/sequential.py in build(self, input_shape)
225 self.inputs = [x]
226 for layer in self._layers:
--> 227 x = layer(x)
228 self.outputs = [x]
229
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/engine/base_layer.py in __call__(self, inputs, **kwargs)
458 # Actually call the layer,
459 # collecting output(s), mask(s), and shape(s).
--> 460 output = self.call(inputs, **kwargs)
461 output_mask = self.compute_mask(inputs, previous_mask)
462
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/layers/wrappers.py in call(self, inputs, training, mask)
246 inner_mask_shape = self._get_shape_tuple((-1,), mask, 2)
247 kwargs['mask'] = K.reshape(mask, inner_mask_shape)
--> 248 y = self.layer.call(inputs, **kwargs)
249 if hasattr(y, '_uses_learning_phase'):
250 uses_learning_phase = y._uses_learning_phase
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/layers/convolutional.py in call(self, inputs)
166 padding=self.padding,
167 data_format=self.data_format,
--> 168 dilation_rate=self.dilation_rate)
169 if self.rank == 3:
170 outputs = K.conv3d(
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in conv2d(x, kernel, strides, padding, data_format, dilation_rate)
3564 strides=strides,
3565 padding=padding,
-> 3566 data_format=tf_data_format)
3567
3568 if data_format == 'channels_first' and tf_data_format == 'NHWC':
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py in convolution(input, filter, padding, strides, dilation_rate, name, data_format)
777 dilation_rate=dilation_rate,
778 name=name,
--> 779 data_format=data_format)
780 return op(input, filter)
781
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py in __init__(self, input_shape, filter_shape, padding, strides, dilation_rate, name, data_format)
854 filter_shape=filter_shape,
855 spatial_dims=spatial_dims,
--> 856 data_format=data_format)
857
858 def _build_op(self, _, padding):
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py in __init__(self, input_shape, dilation_rate, padding, build_op, filter_shape, spatial_dims, data_format)
437 raise ValueError("dilation_rate must be positive")
438 if np.all(const_rate == 1):
--> 439 self.call = build_op(num_spatial_dims, padding)
440 return
441
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py in _build_op(self, _, padding)
863 data_format=self.data_format,
864 strides=self.strides,
--> 865 name=self.name)
866
867 def __call__(self, inp, filter): # pylint: disable=redefined-builtin
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py in __init__(self, input_shape, filter_shape, padding, data_format, strides, name)
134 strides=None,
135 name=None):
--> 136 filter_shape = filter_shape.with_rank(input_shape.ndims)
137 self.padding = padding
138 self.name = name
~/PycharmProjects/Test/venv/lib/python3.6/site-packages/tensorflow/python/framework/tensor_shape.py in with_rank(self, rank)
742 return self.merge_with(unknown_shape(ndims=rank))
743 except ValueError:
--> 744 raise ValueError("Shape %s must have rank %d" % (self, rank))
745
746 def with_rank_at_least(self, rank):
ValueError: Shape (256, 1, 2048, 16) must have rank 3
我完全陷入困境，不知道如何解决问题。
编辑：我用新的错误更新了我的代码，这个错误现在是形状错误。
我相信我找到了解决方案，所以编辑这个答案以反映这一点。根据keras文档，当使用conv2d进行TimeDistribution时，input_shape必须是长度为4的touple
以下是keras文档的一部分。
TimeDistributed可以与任意层一起使用，而不仅仅是Dense，例如Conv2D层：
model = Sequential()
model.add(TimeDistributed(Conv2D(64, (3, 3)),
input_shape=(10, 299, 299, 3)))
对于有类似问题的任何人，我更改了训练和测试数据形状如下：
X_tr = np.expand_dims(np.expand_dims(np.expand_dims(X_train,axis=1),axis=3),axis=1)
X_te = np.expand_dims(np.expand_dims(np.expand_dims(X_test,axis=1),axis=3),axis=1)
和第一层：
model.add(TimeDistributed(Conv2D(kernel_size=(256,1),strides=(32,1),filters=16,activation=LeakyReLU(0.01),data_format='channels_first'),input_shape=(None,1,2048,1)))
它知道运行得很好。