Блог

Я хотел бы создать глубокую сеть, используя две параллельные сети. Одна из них-2D сеть CNN, а другая-LSTM. Входными данными для обеих сетей является двумерный массив чисел с плавающей запятой. Уникальный входной 2D-массив в большую сеть передается каждой из ветвей. Я не знаю, как ввести 2D-массив в ветвь LSTM.

Это мой код:

 def conv_branch(X, F1, s=1):
    
    # Convolutional component 
    X = Conv2D(F1, (3, 3), strides = (s,s), kernel_initializer = glorot_uniform(seed=0))(X)
    
    # Output layer
    X = Flatten()(X)
    X = Dense(F1, activation='relu', kernel_initializer = glorot_uniform(seed=0))(X)
    
    return X

def LSTM_branch(X, F1, s=2):
    
    X = Flatten()(X)
    X = LSTM(32, return_sequences=True)(X)
    X = LSTM(32, return_sequences=True)(X)
    
    # Fully connected component
    X = Dense(F1, activation='relu', kernel_initializer = glorot_uniform(seed=0))(X)

    return X

def concat_stage(X_conv, X_lstm, filters):
    
    # Retrieve Filters
    F1, F2, F3 = filters
    
    # Concatenation layer
    X = Add()([X_conv, X_lstm])

    # Fully connected components
    X = Dense(F1, activation='relu', kernel_initializer = glorot_uniform(seed=0))(X)
    X = Dense(F2, activation='relu', kernel_initializer = glorot_uniform(seed=0))(X)
    X = Dense(F3, activation='relu', kernel_initializer = glorot_uniform(seed=0))(X)

    # Output layer
    X = Dense(4, activation='linear', kernel_initializer = glorot_uniform(seed=0))(X)
    return X

def ConvLSTM(input_shape = (15, 3, 1)):
    
    # Define the input
    X = Input(input_shape)

    # Conv branch
    X_conv = conv_branch(X, 32, s=2)
    
    # LSTM branch
    X_lstm = LSTM_branch(X, 32, s=2)

    # Concatenation
    X_out = concat_stage(X_conv, X_lstm, filters=[64, 32, 16])
    
    # Create model
    model = Model(inputs = X, outputs = X_out)

    return model

model = ConvLSTM()
model.summary()
 

В этом и заключается ошибка:

 ---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-34-5adcf778f629> in <module>
     57     return model
     58 
---> 59 model = ConvLSTM()
     60 model.summary()

<ipython-input-34-5adcf778f629> in ConvLSTM(input_shape)
     47 
     48     # LSTM branch
---> 49     X_lstm = LSTM_branch(X, 32, s=2)
     50 
     51     # Concatenation

<ipython-input-34-5adcf778f629> in LSTM_branch(X, F1, s)
     13 
     14     X = Flatten()(X)
---> 15     X = LSTM(32, return_sequences=True)(X)
     16     X = LSTM(32, return_sequences=True)(X)
     17 

c:python36libsite-packageskeraslayersrecurrent.py in __call__(self, inputs, initial_state, constants, **kwargs)
    498 
    499         if initial_state is None and constants is None:
--> 500             return super(RNN, self).__call__(inputs, **kwargs)
    501 
    502         # If any of `initial_state` or `constants` are specified and are Keras

c:python36libsite-packageskerasenginetopology.py in __call__(self, inputs, **kwargs)
    573                 # Raise exceptions in case the input is not compatible
    574                 # with the input_spec specified in the layer constructor.
--> 575                 self.assert_input_compatibility(inputs)
    576 
    577                 # Collect input shapes to build layer.

c:python36libsite-packageskerasenginetopology.py in assert_input_compatibility(self, inputs)
    472                                      self.name   ': expected ndim='  
    473                                      str(spec.ndim)   ', found ndim='  
--> 474                                      str(K.ndim(x)))
    475             if spec.max_ndim is not None:
    476                 ndim = K.ndim(x)

ValueError: Input 0 is incompatible with layer lstm_8: expected ndim=3, found ndim=2
 

Заранее спасибо за любую помощь