training a network over the CPU

[Yasin ÖZDEMİR]

I want to train a network over the CPU from the games I play over the internet, how can I do this? How can I convert my games to training data and train the network using CPU?

[brian.p.r...@gmail.com]

There are two types of nets typically used with chess programs.

Leela Lc0 nets are large and can be trained with as few as about 40,000 games (for human level strength).

However, GPUs are needed for training as these nets cannot be trained in any reasonable amount of time on CPUs, regardless of how many cores your system may have.

NNUE nets are used with Stockfish and are much smaller and can be trained on CPUs.

Unfortunately, NNUE nets need many more games:  SF uses hundreds of millions to billions of games.

So it is simply not practical to train a net from your games.