如何使用keras序列模型使用gridsearchCV调整l2正则化器

Question

I'm trying to tune the hyperparameter, kernel_regularizer, using gridsearchCV but gridsearchCV keeps telling me that the parameter names I'm entering for kernel_regularizer aren't real parameters

I've tried various parameter names such as l2, kernel_regularizer, kernel, regularizers.l2, regularizers.l2( ) but none have worked.

I've also looked online but can't seem to find any documentation of this issue

My sequential model uses kernel_regularizer=l2(0.01)

param_grid = {'kernel_regularizer': [0.01,0.02,0.03]}

grid = GridSearchCV(...)
grid.fit(x_train, y_train) #this is where I get the error: 
                           #ValueError: kernel is not a legal parameter

Answer 1

You have to wrap your model using KerasClassifier for sklearn GridSearchCV to work.

def get_model(k_reg):
    model = Sequential()
    model.add(Dense(1,activation='sigmoid', kernel_regularizer=k_reg))
    model.compile(loss='binary_crossentropy',optimizer='adam', metrics=['accuracy'])
    return model

param_grid = {
    'k_reg': [ regularizers.l2(0.01), regularizers.l2(0.001), regularizers.l2(0.0001)]
}

my_classifier = KerasClassifier(get_model, batch_size=32)
grid = GridSearchCV(my_classifier, param_grid)

grid.fit(np.random.rand(10,1),np.random.rand(10,1))

Keras Doc , has a detailed example.