Are high n_estimators good?

Thosifer MD 0
Viewed 32

I created a model with CatBoostRegressor. My dataset is 74274 rows × 24 columns.I am using encoding and min-max scaler.

The more I raise the n_estimators value in my model, the better the model score. What is the end of this? How do I decide where to stop? That way I guess it goes forever. Is being high good or bad? Where should the stopping point be?

model = CatBoostRegressor(n_estimators=3000,verbose=False)

model = CatBoostRegressor(n_estimators=10000,verbose=False)

model = CatBoostRegressor(n_estimators=20000,verbose=False)

.

.

.

model regression scikit-learn python
1 Answers

Which set are you checking the score upon?

If that's a train set, the score will likely keep increasing because of overfitting.

For a validation set, the score should stop increasing at a certain point once the order of model complexity becomes comparable to the sample size.

Sklearn/skopt cross validation routines such as GridSearchCV() should aid you in automating the best hyperparameters' selection.

dx2-660
Related