Question

我尝试调整超参数

def create_model():
    # create model
    model = Sequential()
    model.add(Dense(12, input_dim=10, kernel_initializer='uniform', activation='relu'))
    model.add(Dense(4, kernel_initializer='uniform', activation='sigmoid'))
    # Compile model
    model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
    return model


# create model
model = KerasClassifier(build_fn=create_model, verbose=0)
cv = KFold(n_splits=5, shuffle=True, random_state=1)
# grid search epochs, batch size and optimizer
epochs = [ 100, 150, 170, 200, 250]
batches = [ 5, 10, 20, 25, 30]
param_grid = dict(epochs=epochs, batch_size=batches)
grid = GridSearchCV(estimator=model, param_grid=param_grid, cv = cv)

然后适合

%%time
grid_result = grid.fit(X_train, y_train)

我得到了

grid_result.best_params_, grid_result.best_score_
({'batch_size': 10, 'epochs': 150}, 0.31568432165371191)

之后我将完全相同并获得

grid_result.best_params_, grid_result.best_score_
({'batch_size': 20, 'epochs': 100}, 0.31368631761629029)

有什么问题？数据集的大小

X_train.shape, y_train.shape
((1001, 10), (1001, 4))

我认为我们必须使用random_state获得相同的结果，但KerasClassifier()没有random_state

调整Keras的超参数：相同参数的不同结果

0 个答案: