perf: Performance optimizations for samplers

rethinkpriorities / squigglepy

Squiggle programming language for intuitive probabilistic estimation features in Python

MIT License

65 stars 8 forks source link

Closed bcbernardo closed 1 year ago

bcbernardo commented 1 year ago

This PR introduces a few performance optimizations related to distribution sampling. This includes:

Using the a binary search algorithm to search through weights for each simple, instead of the linear search algorithm previously used in the _run_mixture() internal function (samplers.py module).
Using numpy's vectorized functions np.maximum() and np.minimum() to enforce left and right clipping.

bcbernardo commented 1 year ago

@agucova do you want to review it, since you were already working on this?

bcbernardo commented 1 year ago

or @michaeldickens

peterhurford commented 1 year ago

Thanks everyone! This is exciting!