jbirkesteen
commented
2 years ago We want to record winrates and tierates out of 1000 games and save these to a csv with pandas. Forget about illegal rates.
These are the matchups we want:
Let all models play against each other.
Let all models play against random.
Let everyone play against minimax with max_depth=0. Increase depth for the best performers.
We want to let all agents play against each other once we've finished #31 (and maybe determine which one to train even further). Possible suggestion: Wrap a function around testmodel and call this function through a script, looping through relevant models/opponents. note: (remember minimax always needs choices)