search

PathmindAI / nativerl

Train reinforcement learning agents using AnyLogic or Python-based simulations
Apache License 2.0
19 stars 4 forks source link

multiagent balancing #509

Closed thetwotravelers closed 2 years ago

thetwotravelers commented 3 years ago

Small piece of bg_nb that begins to address reward balancing for multiagent models like Felipe's.

slinlee commented 3 years ago

I'm going to test this really quickly on test.devpathmind.com This changes the lines where the previous bug was showing up too.

slinlee commented 2 years ago

Including bg_multi_mini's reward balancing changes made training with reward functions perform worse: https://test.devpathmind.com/sharedExperiment/7273

And using reward terms trains, but also has poor results https://test.devpathmind.com/sharedExperiment/7274

The throughputs are expected to be 60+ but they're in the 10-30s.