Indeed, such chaos can't happen all the time. It's related to the period where JA did nothing. He suddenly came back with the idea to restart compiling engines for Android. Sure, he has capacity and 4 engines per day is a piece of cake.
Nowi i have 77 engines included in the restarting experiment. Be it a RAPIDROID RELOADED, once to have a very KOOOL name and second to reestablish everything to improve the strategy to reach accuracy with minimal effort.
Time control is 10 seconds per move this time, at the quickest acceptable for me. I think i'll never ever go for 5 seconds because below 10 seconds the engines tend to behave somehow unstable. They need more than a blitz timing to warm up and deliver something reflecting their true power. Beyond 10 secs to 15 or 30 or 60, it's much more stable, as per previous experiments.
Last time i'd found 100 games are the minimum threshold to have the whole list stabilized enough, with ELO changes below 10 per 12 more games played. This time, i expect better. We'll see whether i'll come quicker to the target.
500+ games are already played so far in 11 divisions of 7 engines each, playing separately using a thematic opening, which makes 11 different openings per round.
Then, each division exports 2 engines up and 2 engines down while 3 remains in the same division. That looks like a fair exchange ratio in order to uncluster the whole population, to prevent unbalanced pairings and ensure better opponent selection.
I hope to release the first version next week, including only 24 games per engine. Normally, one would predict 24 games are nothing and such ranking is nutz..
I've got a challenging mind and i expect more than that. As promised.
No comments:
Post a Comment