lichess.org
Donate

Deep Learning Self-Play - Bot journey!

Not sure I understood the last iteration of the model. Does it use stockfish to evaluate each possible move from the current position, and then the model chooses between the moves based on their evaluated scores?

Not sure I understood the last iteration of the model. Does it use stockfish to evaluate each possible move from the current position, and then the model chooses between the moves based on their evaluated scores?

@t1u1 said in #2:

Not sure I understood the last iteration of the model. Does it use stockfish to evaluate each possible move from the current position, and then the model chooses between the moves based on their evaluated scores?

For each predicted move of the model, it uses Stockfish just to evaluate CP and then moves the one with the best evaluation.

@t1u1 said in #2: > Not sure I understood the last iteration of the model. Does it use stockfish to evaluate each possible move from the current position, and then the model chooses between the moves based on their evaluated scores? For each predicted move of the model, it uses Stockfish just to evaluate CP and then moves the one with the best evaluation.

About 20 years ago WinBoard(?) had a feature for bots to learn opening statistics from playing against online opponents, and use those statistics to improve against human and bot opponents.

Please do something -- anything -- in the space of adversarial training, to help make something which someday can help humans learn.

About 20 years ago WinBoard(?) had a feature for bots to learn opening statistics from playing against online opponents, and use those statistics to improve against human and bot opponents. Please do something -- anything -- in the space of adversarial training, to help make something which someday can help humans learn.

@Toadofsky said in #6:

Please do something -- anything -- in the space of adversarial training, to help make something which someday can help humans learn.

That's something i really want to do, but it's very hard. I've some ideas anyway, hope i can create something!

@Toadofsky said in #6: > Please do something -- anything -- in the space of adversarial training, to help make something which someday can help humans learn. That's something i really want to do, but it's very hard. I've some ideas anyway, hope i can create something!