使用cross_val_score计算CNN模型交叉验证得分时遇TypeError问题求助

问题场景 #

构建的CNN模型代码如下：

model_cnn = Sequential()
model_cnn.add(Convolution1D(filters=32,kernel_size=2, strides=1, activation='relu', input_shape=(x_train.shape[1], 1)))
model_cnn.add(MaxPooling1D())
model_cnn.add(Flatten())
model_cnn.add(Dense(105))
model_cnn.add(Dropout(0.5))
model_cnn.add(Dense(1, activation="relu"))
model_cnn.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
model_cnn.summary()
x_train = x_train.todense()
x_test = x_test.todense()
x_train = np.expand_dims(x_train, axis=-1)
x_test = np.expand_dims(x_test, axis=-1)
history_cnn = model_cnn.fit(x_train, y_train,validation_data=(x_test, y_test), epochs=3)

执行交叉验证代码时：

print("Mean of Cross validation score(CNN): ", cross_val_score(model_cnn, x_train, y_train, cv=kfold).mean())

出现错误：

TypeError: If no scoring is specified, the estimator passed should have a 'score' method. The estimator <keras.engine.sequential.Sequential object at 0x00000154DF44B6A0> does not.

解决方案 #

方法1：给cross_val_score指定scoring参数 #

Keras的Sequential模型没有默认的score方法，需显式指定评估指标。针对二分类任务，直接指定scoring='accuracy'即可：

print("Mean of Cross validation score(CNN): ", cross_val_score(model_cnn, x_train, y_train, cv=kfold, scoring='accuracy').mean())

方法2：用KerasClassifier包装模型 #

将Keras模型包装为适配scikit-learn接口的KerasClassifier，让模型自动拥有score方法，步骤如下：

1. 导入依赖：

from tensorflow.keras.wrappers.scikit_learn import KerasClassifier

2. 定义模型构建函数：

def build_cnn_model():
    model = Sequential()
    model.add(Convolution1D(filters=32,kernel_size=2, strides=1, activation='relu', input_shape=(x_train.shape[1], 1)))
    model.add(MaxPooling1D())
    model.add(Flatten())
    model.add(Dense(105))
    model.add(Dropout(0.5))
    model.add(Dense(1, activation="sigmoid"))  # 二分类任务建议用sigmoid替代relu
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model

3. 包装模型并执行交叉验证：

model_cnn_wrap = KerasClassifier(build_fn=build_cnn_model, epochs=3, verbose=0)
print("Mean of Cross validation score(CNN): ", cross_val_score(model_cnn_wrap, x_train, y_train, cv=kfold).mean())

额外注意事项 #

原模型最后一层使用relu激活函数，但二分类任务搭配binary_crossentropy损失时，更适合用sigmoid激活——sigmoid输出范围为(0,1)，对应概率值，能更好匹配损失函数的计算逻辑，提升模型效果与评估准确性。

内容的提问来源于stack exchange，提问作者Umer