Closed kepricon closed 2 years ago
@ejunprung the single agent was working already, right? we need the multi agent to work now?
@slinlee It's not related to single vs multi-agent. I was testing the gym environment versus Pathmind simulation environment. I just happened to have the multi-mouse example ready to go at the time so I was using that for testing. It should still work fine, I haven't see any issues around multi-agent so far.
@ejunprung so what works right now?
Still figuring that out. Need another day or two to finish testing so I'll compile a list after that. But so far, gym works fine. Pathmind simulations is broken but I still need to test Dae's fix.
Multi-agent mechanically should work but it's missing features (e.g. skip) so it won't be usable in practice.
k. yeah nice, def keep a list of the small things that need to be added
run tests
@slinlee I think we need to add this fix to test environment. I still can't get my local py-nativerl working correctly. Do I need to build a new NativeRL or is that automated now?
I tested with Ed's model(added getRewardTerms() into Mdoe's Simulation.py) model_py_simulation.zip
Here are Training results
https://s3.console.aws.amazon.com/s3/buckets/dh-training-dynamic-files.pathmind.com?region=us-east-1&prefix=id3086/output/&showversions=false