LogisticRegression spends a lot of time in exp and log1p
Open
- Dominant language
- Python
- Stars
- 951
- Forks
- 262
- PR merge metrics
- No merged PRs in 30d
Description
Something like 200s in comparison to the 13s spent in `.dot` / matvec. This is from looking at the profiler after running through the dask-glm example notebook.
Is it normal to spend so much time in scalar functions? Are there approximations that make sense to do?
This is low priority from my perspective (there is plenty more to do) but I thought I'd record it.
Contributor guide
Assessment
This issue has not been assessed yet.