连接多个卷积块：“应在至少 2 个输入的列表上调用“连接”层”

Question

我正在尝试重新创建一篇论文中提到的卷积模型，但我遇到了一些问题。

这是论文中模型的架构：

这是我对 Python 和 Tf 的尝试：

X = df
y = dataset['attack_map']

X_train, X_test, y_train, y_test = train_test_split(X,y, test_size = 0.30)
X_train = np.array(X_train)
X_test = np.array(X_test)
y_train = np.array(y_train)
y_test = np.array(y_test)


X_train = X_train.reshape(X_train.shape[0], X_train.shape[1], 1)
X_test = X_test.reshape(X_test.shape[0], X_test.shape[1], 1)

shape = (X_train.shape[1], X_train.shape[2])


## FIRST BLOCK
model1 = tf.keras.Sequential()
model1.add(tf.keras.layers.InputLayer(input_shape = shape))
model1.add(tf.keras.layers.Conv1D(filters = 32, kernel_size = 2, strides = 1))
model1.add(tf.keras.layers.Activation('relu'))
model1.add(tf.keras.layers.MaxPooling1D(pool_size = 2, strides = 1))
model1.add(tf.keras.layers.Flatten())
model1.add(tf.keras.layers.Dropout(.5))

## SECOND BLOCK
model2 = tf.keras.Sequential()
model2.add(tf.keras.layers.InputLayer(input_shape = shape))
model2.add(tf.keras.layers.Conv1D(filters = 32, kernel_size = 4, strides = 1))
model2.add(tf.keras.layers.Activation('relu'))
model2.add(tf.keras.layers.MaxPooling1D(pool_size = 4, strides = 1))
model2.add(tf.keras.layers.Flatten())
model2.add(tf.keras.layers.Dropout(.5))

## THIRD BLOCK
model3 = tf.keras.Sequential()
model3.add(tf.keras.layers.InputLayer(input_shape = shape))
model3.add(tf.keras.layers.Conv1D(filters = 32, kernel_size = 8, strides = 1))
model3.add(tf.keras.layers.Activation('relu'))
model3.add(tf.keras.layers.MaxPooling1D(pool_size = 8, strides = 1))
model3.add(tf.keras.layers.Flatten())
model3.add(tf.keras.layers.Dropout(.5))
model3.add(tf.keras.layers.Dense(256))
model3.add(tf.keras.layers.Dropout(.5))
model3.add(tf.keras.layers.Dense(5))
model3.add(tf.keras.layers.Softmax())


model = tf.keras.Sequential()      
model.add(tf.keras.layers.Concatenate([model1, model2, model3]))
model.add(tf.keras.layers.Dense(5))
model.add(tf.keras.layers.Softmax())



model.compile(loss = 'sparse_categorical_crossentropy', optimizer = "adam", metrics = ['accuracy'])
callback = tf.keras.callbacks.EarlyStopping(monitor="val_loss", min_delta=0, patience=10, verbose=1, mode="auto", baseline=None, restore_best_weights=False),


start = time.perf_counter()
model.fit(X_train, y_train, epochs=EPOCHS, callbacks=[callback])
elapsed = time.perf_counter() - start


model.evaluate(X_test,  y_test, verbose=2)
print('Elapsed %.3f seconds.' % elapsed)

我得到的错误是：

 ValueError: A `Concatenate` layer should be called on a list of at least 2 inputs

所以想法是连接这些块。这可能是一个愚蠢的问题，但我最近一直在使用 tf。

有什么建议吗？谢谢

Answer 1

您应该在 keras 中使用函数式 API。通过这样做，您可以使用以下伪代码轻松实现它：

h1=tf.keras.layers.Conv1D(filters = 32, kernel_size = 4, strides = 1)(inp)
h1=tf.keras.layers.Activation('relu')(h)
h1=tf.keras.layers.Flatten()(h)
h1=tf.keras.layers.Dense(5)(h)
out1=tf.keras.layers.Softmax()(h)
...
...
...
h2=tf.keras.layers.Dense(5)(h)
out2=tf.keras.layers.Softmax()(h)
...
...
...
h3=tf.keras.layers.Dense(5)(h)
out3=tf.keras.layers.Softmax()(h)

基本上输出值表示不同结果密集层的输出。（模型）

--

h=tf.keras.layers.concatenate([out1,out2,out3])
h=tf.keras.layers.Dense(5)(h)
R_out=tf.keras.layers.Softmax()(h)
model=tf.keras.Model(inputs=inp,outputs=R_out)