A simple engine using NEAT (NeuroEvolution of Augmenting Topologies)
Posted: Fri Dec 03, 2021 9:53 am
Hi everyone!
I havnt found any implementation of a NEAT chess engine and I'm curious to see how (or if) it performs so I'm planning to code something using NEAT-python and python-chess libraries.
I was thinking starting from a network with 12 x 64 + 4 input cells (12 planes of 8x8 bits to encode pieces positions and 4 bits to track castles for each player) and one output cell to rate the position from -1 to 1 and selecting the best legal move on each turn. Maybe tracking the ELO of the population and select the top 20% for reproduction at each generation.
Have anyone here ever tried something like that? Or have any suggestions?
Thanks for your help!
I havnt found any implementation of a NEAT chess engine and I'm curious to see how (or if) it performs so I'm planning to code something using NEAT-python and python-chess libraries.
I was thinking starting from a network with 12 x 64 + 4 input cells (12 planes of 8x8 bits to encode pieces positions and 4 bits to track castles for each player) and one output cell to rate the position from -1 to 1 and selecting the best legal move on each turn. Maybe tracking the ELO of the population and select the top 20% for reproduction at each generation.
Have anyone here ever tried something like that? Or have any suggestions?
Thanks for your help!