search

ogrisel / pygbm

Experimental Gradient Boosting Machines in Python with numba.
MIT License
183 stars 32 forks source link

Use cache=True on everything? #66

Open NicolasHug opened 5 years ago

NicolasHug commented 5 years ago

Opened https://github.com/numba/numba/issues/3588 to ask if there's an alternative.

NicolasHug commented 5 years ago

There's none (for now).

Do we want to cache all compilations @ogrisel ? I'd say yes, since it takes a non-negligible amount of time

NicolasHug commented 5 years ago

So I did that, cache=True is not supported for jitclasses nor when parallel=True so gain is pretty minimal: going from about 10s to 9s.

ogrisel commented 5 years ago

Ok so better wait for a better caching infrastructure in numba before enabling it.