HOTFIX - Fix metrics bug

PathmindAI / nativerl

Train reinforcement learning agents using AnyLogic or Python-based simulations

Apache License 2.0

19 stars 4 forks source link

HOTFIX - Fix metrics bug #499

Closed slinlee closed 3 years ago

slinlee commented 3 years ago

there's a bad discrepancy in simulation metrics between NativeRL 1.8.0 --> 1.8.1

1.8.0 --> https://app.pathmind.com/sharedExperiment/17922

1.8.1 --> https://app.pathmind.com/sharedExperiment/ in 1.8.1, the simulation metrics seem to be divided by 2. they are half what they should be

Introduced in #441. This was setting the metrics back to zero with each agent.

maxpumperla commented 3 years ago

hmm, I feel like I touched this exact line a while ago in that bg_simba branch, along with a few other things. where did that PR go? did we split it up?

slinlee commented 3 years ago

hmm, I feel like I touched this exact line a while ago in that bg_simba branch, along with a few other things. where did that PR go? did we split it up?

@maxpumperla you're right, I just found it's fixed in dev. https://github.com/SkymindIO/nativerl/compare/prod...dev#diff-663a64ccb155bc9c502f79b94d8124001413e6752b9d517c3d86058776790e47R322-R323

slinlee commented 3 years ago

tests passed here https://jenkins.dev.devpathmind.com/blue/organizations/jenkins/Nativerrl_PRs/detail/Nativerrl_PRs/752/pipeline