dmlc/tl2cgen

Disable missing value check

Opened this issue · 3 comments

cckao commented

Hi,

Is it possible to disable missing value check in generated codes? In many cases this is unnecessary and I think disabling it would improve prediction speed. Thanks a lot.

Would be nice for sure!

I'm using a regex to get around of it instead:

!\(data\[[0-9]*\].missing.!=.-1\).\|\|

hcho3 commented

I'll add an option to disable missing value checks. Some models do not use missing values, unlike XGBoost.

Would very much appreciate a missing value flag.