0

hyperopt用来寻找catboost回归量的最佳超参数。我正在遵循本指南。相关部分是:


ctb_reg_params = {
    'learning_rate':     hp.choice('learning_rate',     np.arange(0.05, 0.31, 0.05)),
}
ctb_fit_params = {
    'verbose': False
}
ctb_para = dict()
ctb_para['reg_params'] = ctb_reg_params
ctb_para['fit_params'] = ctb_fit_params
ctb_para['loss_func' ] = lambda y, pred: np.sqrt(mean_squared_error(y, pred))

def ctb_reg(self, para):
    reg = ctb.CatBoostRegressor(**para['reg_params'])
    reg.fit(x_train, y_train, **para['fit_params'])
    pred = reg.predict(x_test)
    loss = para['loss_func'](y_test, pred)
    return {'loss': loss, 'status': STATUS_OK}

fmin(fn=ctb_reg, space=ctb_para, algo=tpe.suggest, max_evals=100, trials=Trials())

几分钟后,我得到了这个:

{'learning_rate': 4}

如何提取最佳学习率?是 np.arange(0.05, 0.31, 0.05)[4]吗?有没有更好的提取方法?

4

1 回答 1

1
from hyperopt import space_eval
print(space_eval(ctb_para, fmin_result))
于 2020-01-23T12:30:25.943 回答