Yes, the file size actually grows. But what are the real benefits of all this? After 300 5 minute blitz games I didn't notice any improvement in performance. Why?
■ Here are some questions and answers put to another group:
I have some questions about the capability who have many Stockfish derivate versions like Shashchess and Sugar NN to learn.
1) They learn through self play games like Lc0 ?
Or after a series of games against other engines ?
2) Can they learn via a human game database ?
3) If i set the limit strenght option i lose the ability of learning ?
4) After how many games an engine can learn something different ?
5) Is there an ability to make a weaker and more human like engine after the learning process ?
6) What are the steps i need to follow to make engines like Sugar or Shashchess ready for learning ?
Thanks for your reply...
1. They can learn playing against himself and against other engines
4. At long time controls, AT LEAST 11 GAMES.
At bullet (1m/game), after AT LEAST 500 GAMES.
5. Yes, combining learning with handicp mode
6. See the Readme associated to ShashChess"