import catboost as cb
import lightgbm as lg

from sklearn.datasets import make_regression
X, Y = make_regression(n_samples=500_000, n_features=100)
Y = (Y - Y.mean()) / Y.std()


class MyLoss(object):
    def calc_ders_range(self, approxes, targets, weights):
        ders = np.zeros((len(targets), 2))
        ders[:, 0] = targets - approxes
        ders[:, 1] = -1
        return ders

    def evaluate(self, approxes, target, weight):
        return np.mean((approxes[0] -  target)**2), 1

    def is_max_optimal(self):
        True

    def get_final_error(self, error, weight):
        return error

    def lg_loss(self, target, approxes):
        '''for lightgbm test'''
        ders = self.calc_ders_range(approxes, target, None)
        return ders[:, 0], ders[:, 1]


lg_reg = lg.LGBMRegressor(objective=MyLoss().lg_loss, max_depth=6, n_estimators=10, learning_rate=0.1, verbose=1)
lg_reg.fit(X, Y)

reg = cb.CatBoostRegressor(max_depth=6, n_estimators=10, learning_rate=0.1)
reg.fit(X, Y)

reg2 = cb.CatBoostRegressor(objective=MyLoss(), eval_metric=MyLoss(), max_depth=6, n_estimators=10, learning_rate=0.1)
reg2.fit(X, Y)

# Test
pred = reg.predict(X)
pred2 = reg2.predict(X)
print(f"max_diff = {np.max(np.abs(pred - pred2)) }")