Dataset size and learning rate #2573
denisvorotyntsevbidease
started this conversation in
General
Replies: 1 comment
-
it depends but you may have a look at how catboost chooses learning rate |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Problem:
I train a CatBoost model using the training part of the dataset to estimate the validation error. Then, I retrain the model using the full data (training + validation). Do I need to change the hyperparameters of CatBoost to improve performance (number of trees, learning rate, etc.)?
I heard a recommendation to increase the learning rate by the square root of the increased sample size (e.g., 30% more data -> increase the learning rate by 1.3^0.5), but I haven't found any theoretical justifications yet. Any recommendations? Is it a heuristic?
catboost version:
Operating System:
CPU:
GPU:
Beta Was this translation helpful? Give feedback.
All reactions