合并多个CNN

Question

我正在尝试对Conv1D模型中的多个输入执行。所以我有 15 个大小为 1x1500 的输入，其中每个输入都是一系列层的输入。所以我有 15 个卷积模型，我想在全连接层之前合并它们。我已经在一个函数中定义了卷积模型，但是我无法理解如何调用该函数然后将它们合并。def defineModel(nkernels, nstrides, dropout, input_shape):    model = Sequential()    model.add(Conv1D(nkernels, nstrides, activation='relu', input_shape=input_shape))    model.add(Conv1D(nkernels*2, nstrides, activation='relu'))    model.add(BatchNormalization())    model.add(MaxPooling1D(nstrides))    model.add(Dropout(dropout))    return modelmodels = {}for i in range(15):    models[i] = defineModel(64,2,0.75,(64,1))我已经成功地连接了 4 个模型，如下所示：merged = Concatenate()([ model1.output, model2.output, model3.output, model4.output])merged = Dense(512, activation='relu')(merged)merged = Dropout(0.75)(merged)merged = Dense(1024, activation='relu')(merged)merged = Dropout(0.75)(merged)merged = Dense(40, activation='softmax')(merged)model = Model(inputs=[model1.input, model2.input, model3.input, model4.input], outputs=merged)由于单独编写 15 层效率不高，我如何在 for 循环中为 15 层执行此操作？

Answer 1

我认为你能做的最好的事情就是在任何地方使用函数式 API：

def defineModel(nkernels, nstrides, dropout, input_shape):

    l_input = Input( shape=input_shape )

    model = Conv1D(nkernels, nstrides, activation='relu')(l_input)

    model = Conv1D(nkernels*2, nstrides, activation='relu')(model)

    model = BatchNormalization()(model)

    model = MaxPooling1D(nstrides)(model)

    model = Dropout(dropout)(model)

    return model, l_input

models = []

inputs = []

for i in range(15):

    model, input = defineModel(64,2,0.75,(64,1))

    models.append( model )

    inputs.append( input )

然后很容易恢复子模型的输入和输出列表并合并它们

merged = Concatenate()(models)

merged = Dense(512, activation='relu')(merged)

merged = Dropout(0.75)(merged)

merged = Dense(1024, activation='relu')(merged)

merged = Dropout(0.75)(merged)

merged = Dense(40, activation='softmax')(merged)

model = Model(inputs=inputs, outputs=merged)

通常，这些操作不是瓶颈。这些都不应该在训练或推理过程中产生重大影响

Answer 2

当然，正如@GabrielM 建议的那样，使用函数式 API 是最好的方法，但是如果你不想修改define_model函数，你也可以这样做：

models = []

inputs = []

outputs = []

for i in range(15):

    model = defineModel(64,2,0.75,(64,1))

    models.append(model)

    inputs.append(model.input)

    outputs.append(model.output)

merged = Concatenate()(outputs) # this should be output tensors and not models

# the rest is the same ...

model = Model(inputs=inputs, outputs=merged)