Closed JaysenStark closed 6 years ago
the random situations are brute force solved on a supercomputer to initialise the neural network.
It's the first scenario. As you correctly describe, we use SOLVING algorithm to generate data and train net. This net is then used during RESOLVING
Hi, i am a student from HIT focus on imperfect information game. I noticed that training data are generated by solving random poker situation, and in this project, the solving algorithm is re-sloving, but re-solving itself needs support from neural network which are trained with training data. I am confused. Could I understand this way: you use one sloving algorithm to solve random poker situation to generate data, then train value network; when it comes to play texas hold'em, you use another solving method (re-solving) to do real time computation based on values given by value network ? Or you use re-solving to solve RIVER round random poker situation, then construct RIVER round value network, then use re-solving to solve TURN round random poker situation based on previously built value network, then construct TURN value network, then flop...then preflop? I am looking forward to your reply.