Introduce a `play` run to execute (and observe) a few trials using any actor implementation

cogment / cogment-verse

Research platform for Human-in-the-loop learning (HILL) & Multi-Agent Reinforcement Learning (MARL)

https://cogment.ai/cogment_verse

Apache License 2.0

78 stars 14 forks source link

Closed cloderic closed 2 years ago

cloderic commented 2 years ago

This PR is built upon #36 which should be integrated first

closes #11

Also:

Introduce a random actor implementation. Only works with discrete action spaces for now
Introduce the concept of EnvironmentSpec as a datastructure communicating the characteristics of an environment between runs and actors.
Models are serialized alongside their environment specs to be easier to reload and check.
Introduce a role for the human player, disable webclient controls when as observer

cloderic commented 2 years ago

A followup concern will be splitting of the run_params.yml which become a little unwieldy, it's gonna be tackled in here #49