Skip to content

ovl parallelization

Reed Essick requested to merge ovl-parallelization into master

this parallelizes the evaluation of configuration performance within OVL, DOVL, and DOVLfap. Specifically, we modify _recalculate to condition on whether incremental>0 in order to evaluate the result serially. If incremental==0, we use multiprocessing.pool.map_async to distribute the evaluation of configuration performances to multiple cores. This should hopefully speed up the initial vetolist evaluation, which currenlty dominates OVL's training runtime.

Fixes #89 (closed).

Edited by Reed Essick

Merge request reports