python - ValueError：检查目标时出错：预期dense_1的形状为（16、1、2）但得到的数组形状为（2、1、1）

Question

我对使用机器学习和 keras 很陌生。这是我的第一个 CNN 模型。我建立了一个模型来将声音分类到 UrbanSound8kDataset 中。当我在一个小的虚拟集上运行模型时，我不断收到以下错误：

ValueError: Error when checking target: expected dense_1 to have shape (16, 1, 2) but got array with shape (2, 1, 1)

我查看了其他有类似问题的帖子，并尝试将模型损失更改为 sparse_categorical_crossentropy 而不是 categorical_crossentropy。我还在第一个密集层之前添加了 model.flatten() 。不幸的是，这些都没有解决我的错误。

这是我的模型的相关代码：

num_labels = yy.shape[1]
filter_size = 2
num_rows = 1
num_columns = 1
num_channels = 1    

x_train = x_train.reshape(x_train.shape[0], num_rows, num_columns, num_channels)
y_train = y_train.reshape(y_train.shape[0], 2, num_columns, num_channels)
x_test = x_test.reshape(x_test.shape[0], 1, num_columns, num_channels)
y_test = y_test.reshape(y_test.shape[0], 2, num_columns, num_channels)

model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, input_shape=(num_rows,

    num_columns, num_channels), activation='relu', data_format='channels_first',   padding="same"))
model.add(MaxPooling2D(pool_size=1))
model.add(Dropout(0.2))

model.add(Conv2D(filters=32, kernel_size=1, activation='relu'))
model.add(MaxPooling2D(pool_size=1))
model.add(Dropout(0.2))

model.add(Conv2D(filters=64, kernel_size=1, activation='relu'))
model.add(MaxPooling2D(pool_size=1))
model.add(Dropout(0.2))

model.add(Conv2D(filters=64, kernel_size=1, activation='relu'))
model.add(MaxPooling2D(pool_size=1))
model.add(Dropout(0.2))

#Flatten layer here:

model.add(Dense(num_labels, activation='softmax'))
model.compile(loss='categorical_crossentropy', metrics=['accuracy'], optimizer='adam')

model.summary()


num_epochs = 100
num_batch_size = 132

start = datetime.now()



model.fit(x_train, y_train, batch_size=num_batch_size, epochs=num_epochs, validation_data=(x_test, y_test), verbose=1)

以下是模型摘要：

Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 16, 1, 1)          80        
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 16, 1, 1)          0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 16, 1, 1)          0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 16, 1, 32)         64        
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 16, 1, 32)         0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 16, 1, 32)         0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 16, 1, 64)         2112      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 16, 1, 64)         0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 16, 1, 64)         0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 16, 1, 64)         4160      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 16, 1, 64)         0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 16, 1, 64)         0         
_________________________________________________________________
dense_1 (Dense)              (None, 16, 1, 2)          130       
=================================================================
Total params: 6,546
Trainable params: 6,546
Non-trainable params: 0
____________________________

错误发生在我总结的代码的最后一行，model.fit 行。

如果我需要包含更多代码，请告诉我。这只是我第二次在这里发帖，所以我不确定要包含多少代码。

非常感谢！

标签： pythonmachine-learningkeras