分类和回归的组合

from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dense, concatenate

# 假设输入特征维度为10
input_layer = Input(shape=(10,))

# 共享的特征提取层
shared_layer = Dense(64, activation='relu')(input_layer)

# 分类任务的输出层
classification_output = Dense(num_classes, activation='softmax', name='classification')(shared_layer)

# 回归任务的输出层
regression_output = Dense(1, activation='linear', name='regression')(shared_layer)

# 构建模型
model = Model(inputs=input_layer, outputs=[classification_output, regression_output])

# 编译模型，为两个任务设置不同的损失函数
model.compile(optimizer='adam',
              loss={'classification': 'categorical_crossentropy', 'regression': 'mean_squared_error'},
              metrics={'classification': 'accuracy', 'regression': 'mae'})

# 假设X_train和y_train分别是训练数据和标签
# y_train包含分类标签和回归目标值
model.fit(X_train, {'classification': y_train_classification, 'regression': y_train_regression}, epochs=10)