Smaller dataset takes 10 times longer for training

[JinwooMcLee]

First of all, thank you for publicize this amaizing repo.

Recently I'm working with MARS model. All things works great, but there's one thing confuses me.

When train MARS with whole dataset (Rows, Columns) = (630K, 69), it takes 1 to 2 minutes. But when I train MARS with subset of that dataset (430K, 69), it takes 14 to 15 minutes. Also, Use_fast doesn't affect much with training time with my dataset.

I'm wondering if there's any way to reduce subset training time.

Just to let you know,

• Whole set y ranges form 175 ~ 220
• Subset y ranges from 200 ~ 220

Thank you for all your effort. Sincerely.