How does max_bin param affect the speed and overfitting of LightGBM?

Question

This is from lightgbm doc But I still don't understand what is bin? and how it is affecting the performance of lightgbm?

Answer 1

A key part of LightGBM is that it bins continuous features into bins/buckets (like a histogram) when building trees, which reduces memory use and massively speeds up training time. See section 4.1 of the docs, here.

XGBoost has now implemented this as an option too, which allows similar training speeds as LightGBM and can be implemented using tree_method='hist'.