Light GBM parameter tuning

Introduction

In this article, we will summarize the parameters of LightGBM. I have just translated the Official Document into Japanese, but I hope you find it helpful. (I would like to update the story of LightGBM itself from time to time.)

Three important parameters

max_depth Specifies the depth of the tree to use. Let's think about it with the parameter num_leaves.

num_leaves It is a parameter that determines the complexity of the model. Theoretically, 2 ^ (max_depth) seems to be good, but in practice, a value smaller than 2 ^ (max_depth) seems to be good. ** If max_depth is specified as 7, should num_leaves be specified as about 70-80? Was written. ** **

min_data_in_leaf It is an important parameter to prevent overfitting. It is written that it is decided by the number of samples of training data and the value of num_leaves. When dealing with a large number of samples, it seems that a value of hundreds to thousands is good.

Finally

I found a neatly organized site. Please read this as well!

Recommended Posts

Light GBM parameter tuning
Parameter tuning with luigi (2)
Parameter tuning with luigi