Posts tagged reinforcement-learning
- The best truco agent I trained was not the last one
I trained a Truco Gaúcho agent by self-play for 5 million steps. It reached 73% against a rule-based bot and ended at 63%, and the training metric showed none of it.