what are the best performing models?

[shouldsee]

Thanks for sharing! Just wondering what's going on in terms of

• The current workflow to implement and test new models?
• what metrics to calculate for new models?
• what are the best models so far?

[Cybernetic1]

Hey, sorry I did not see your message.

1. First we find an algorithm (such as PPO or Soft Actor-Critic) and adapt its code to solve Tic Tac Toe. The first version would use board-vector and second version use logic proposition embedding.
2. For each game the highest score is 20. I just want to demonstrate convergence first.
3. All the models tested so far -- based on naive policy gradient -- failed to converge to the highest score. Some of them reached close to the highest score but were unstable.

[shouldsee]

Ok thanks. I guess the core question is, how is the score calculated? which function did you use to calculate the score?