ValueError：检查输入时出错：预期 conv1d_29_input 有 3 个维度，但得到了形状为 (150, 1320) 的数组

Question

momentum_rate = 0.5
learning_rate = 0.1
neurons = 30

def convolutional_neural_network(x, y):
    print("Hyper-parameter values:\n")
    print('Momentum Rate =',momentum_rate,'\n')
    print('learning rate =',learning_rate,'\n')
    print('Number of neurons =',neurons,'\n')
    model = Sequential()
    #model.summary()
    model.add(Conv1D(input_shape=(X.shape[1],X.shape[0]),activation='relu',kernel_size = 1,filters = 64))
    
    model.add(Flatten())
    
    model.add(Dense(neurons,activation='relu')) # first hidden layer
    model.summary()
    model.add(Dense(neurons, activation='relu'))
    model.summary()# second hidden layer
    model.add(Dense(neurons, activation='relu'))
    model.summary()
    model.add(Dense(neurons, activation='relu'))
    model.summary()
    model.add(Dense(10, activation='softmax'))
    model.summary()
    sgd = optimizers.SGD(lr=learning_rate, decay=1e-6, momentum=momentum_rate, nesterov=True)
    model.summary()
    model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy',tensorflow.keras.metrics.Precision()])
    model.summary()
    history = model.fit(X, y, validation_split=0.2, epochs=10)
    model.summary()
    print("\nTraining Data Statistics:\n")
    print("CNN Model with Relu Hidden Units and Cross-Entropy Error Function:")

print(convolutional_neural_network(X,y))

X 的形状是 (150, 1320) y 的形状是 (150,)

这是我得到的输出：

Model: "sequential_36"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d_30 (Conv1D)           (None, 1320, 64)          9664      
_________________________________________________________________
flatten_21 (Flatten)         (None, 84480)             0         
_________________________________________________________________
dense_106 (Dense)            (None, 30)                2534430   
_________________________________________________________________
dense_107 (Dense)            (None, 30)                930       
_________________________________________________________________
dense_108 (Dense)            (None, 30)                930       
_________________________________________________________________
dense_109 (Dense)            (None, 30)                930       
_________________________________________________________________
dense_110 (Dense)            (None, 10)                310       
=================================================================
Total params: 2,547,194
Trainable params: 2,547,194
Non-trainable params: 0

ValueError: Error when checking input: expected conv1d_30_input to have 3 dimensions, but got array with shape (150, 1320)

Answer 1

Conv1D需要一个input_shape形式的(steps, input_dim) （ 参见文档）。 现在，如果我正确理解您的input_dim=1因为 1320 是样本数，150 是数组的长度。 在这种情况下，更改input_shape=(X.shape[1], X.shape[2]) 。

编辑：目前还不清楚你想做什么。 下面的代码正在运行，并显示了您网络的预期形状。 但请注意，我更改了 y 维度以匹配行数和输出层。 我不确定 y 形状 (150,) 代表什么。

X = tf.random.normal((1320,150,1))
y = tf.random.uniform((1320,10))

momentum_rate = 0.5
learning_rate = 0.1
neurons = 30

def convolutional_neural_network(x, y):
    print("Hyper-parameter values:\n")
    print('Momentum Rate =',momentum_rate,'\n')
    print('learning rate =',learning_rate,'\n')
    print('Number of neurons =',neurons,'\n')
    model = Sequential()
    #model.summary()
    model.add(Conv1D(input_shape=(X.shape[1], X.shape[2]),activation='relu',kernel_size = 1,filters = 64))
    
    model.add(Flatten())
    
    model.add(Dense(neurons,activation='relu')) # first hidden layer
    model.add(Dense(neurons, activation='relu'))
    model.add(Dense(neurons, activation='relu'))
    model.add(Dense(neurons, activation='relu'))
    model.add(Dense(10, activation='softmax'))
    sgd = optimizers.SGD(lr=learning_rate, decay=1e-6, momentum=momentum_rate, nesterov=True)
    model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'] )
    history = model.fit(X, y, validation_split=0.2, epochs=10)
    model.summary()
    print("\nTraining Data Statistics:\n")
    print("CNN Model with Relu Hidden Units and Cross-Entropy Error Function:")

print(convolutional_neural_network(X,y))

Answer 2

由于您的错误反映了您的输入形状是(150, 1320) 。 在评论中你说你有 1320 个样本（行）和 150 个特征（列）。

让我们制作一些具有上述形状的临时数据作为X和y ：

X = tf.random.uniform((150,1320))
y = tf.random.uniform((1320,10))  
#10 label for each sample which maybe a little strange, take care of it

现在我们有X与形状(150,1320)和y与形状(1320,10)

由于我们有 1320 个样本，它应该是第一个轴，我们必须将它转置：

X = tf.transpose(X)

现在 X 形状将是(1320,150)而不是(150,1320) 。

由于Conv1D 层需要输入为batch_shape + (steps, input_dim) ，我们需要添加一个新维度。 所以：

X = tf.expand_dims(X,axis=2)
print(X.shape, y.shape)     # X.shape=(1320, 150, 1) y.shape=(1320,10)

然后，我们的 X 形状为(1320,150,1)

现在，让我们在Conv1D层中指定输入形状：

model.add(Conv1D(input_shape=(X.shape[1:]),activation='relu',kernel_size = 1,filters = 64))