Which loss function shoud be chosen when use eval_metric in CatBoostRegressor?

Question

I use this code to test my dataset with CatBoostRegressor:

import pandas as pd
from sklearn.model_selection import train_test_split
import catboost

train_dat