LightGBM icon indicating copy to clipboard operation
LightGBM copied to clipboard
verified publisher icon microsoft

GOSS Sampling

Open louise-vuitton2203 opened this issue 7 months ago • 0 comments

Hi,

I am new to LightGBM, and I was reading about GOSS. Please help me understand the following questions:

  1. From the Algorithm description, it said that sample B (sample with small gradient) has a size of b x len(I) but in page 4, the article mentioned 'it randomly samples b × 100% instances from the rest of the data'. So from what I understand here, sample B should has a size of b x len((1-a)*I). So which one is the correct answer? Image

  2. I remember reading that the data points are sorted according to their gradients and hessian values for GOSS. Why does the algorithm, and by extension, the whole paper,r not mention Hessian?

louise-vuitton2203 avatar May 11 '25 07:05 louise-vuitton2203