bernds wrote:
I was curious, so I was trying to run something similar yesterday. While I got a similar result, the games all looked identical up to move 40 or so.
I think I saw a comment in the github by someone else who ran comparisons that one of the tools people sometimes use to run these head to head matches (twogtp?) doesn't restart the engine between games, it just uses the existing engine in memory and sets up the new board/etc and begins playing the new game. The issue with this at the time (perhaps a recent pull request addressed it) is that the NN eval cache doesn't flush just because you cleared the board. Thus you'll have exactly the same results of every early NN eval each game you play. (Normally symmetry causes some small random noise, but once cached the value never changes.)