McBrain 2017 2.4 - an early release!

[MikeB]

Hi

Thanks for this new McBrain, I am always interested to play against these engines and to test the limit ratings interpretations. For me style and 'Human like play' is important as apparently this isnt always easy to implement when an engine is hadicapped in some way. Your stockfish derivatives seem to do a good job in my opinion on this score.

I am not a strong player, Kasparov's underpants would probably beat me in a game of chess, so for me, absolute strength isnt an issue. There is always the latest Stockfish etc.

If I may, a quick question- I read somewhere about the random factor for the limit ratings, is that built in to limit ratings or is it an option under 'variety' ?

Anyway ,thanks for your efforts.

NR

If you use the limiti strength , each rating (1101 in total -1725 to 2825) is tied a specific nodes per second (nps) starting at the lowest rating of 1725 which has a target of 32 nps. 1726 will be slightly higher and so when you get to 1825 the target is 64 nps. It doubles every 100 ELO , but it does so at steady rate of approximately 1.007 times the previous ELO target. 1.007^100 = 2 , hence the nps doubles every 100 ELO which is accepted by many as the approximate ELO when you double the speed of an engine - but each engine is different so it may not be applicable to all engines. The random factor stores the ELO you selected and on each and every move it will randomly adjust that ELO up for that move. It is a range of -33 ELO to + ELO a range of 66 ELO. A change of +33 ELO will increase the nps by ~25% - so the playing level for each every move will be different - the inconsistency is more human like and it will be slightly harder to get repeat games while playing at one ELO level. I appreciate your comments !

[MikeB]

Lydumil,

Fair enough. One quick comment regarding "cherry picking " the results and showing only the positive results. That is true , but not for the reason why you think. I run many , many tests and most of them fail miserablely - why would anyone be interested in those ? So of course I do not post failed results - I may posted but not very much. I think the real issue is two fold , I get perhaps get over exuberant on a positive result that in the end , to your point , ultimately did not pan out - and I'm the type of person who looks at empty glass and see something good in it. In all sincerity , I barely run enough games for me or anyone else to make an assessment one way or the other - so I need to change how I present those results so folks such as yourself and others do not think I'm shamelessly touting McBrain on an unsupported basis. I will take that as an action item on my end and if I miss the mark feel free to let me know diplomatically how I missed the mark and messed up.
Here's to a new beginning! 👍😊

[Gusev]

An earlier version of McBrain is doing reasonably well at the classic time control so far, http://www.talkchess.com/forum/viewtopi ... 60&t=63781.

[fantasmadel50]

Lydumil,

Fair enough. One quick comment regarding "cherry picking " the results and showing only the positive results. That is true , but not for the reason why you think. I run many , many tests and most of them fail miserablely - why would anyone be interested in those ? So of course I do not post failed results - I may posted but not very much. I think the real issue is two fold , I get perhaps get over exuberant on a positive result that in the end , to your point , ultimately did not pan out - and I'm the type of person who looks at empty glass and see something good in it. In all sincerity , I barely run enough games for me or anyone else to make an assessment one way or the other - so I need to change how I present those results so folks such as yourself and others do not think I'm shamelessly touting McBrain on an unsupported basis. I will take that as an action item on my end and if I miss the mark feel free to let me know diplomatically how I missed the mark and messed up.
Here's to a new beginning! 👍😊

I congratulate him for his work and dedication, Mc brain 2.4 is a great engine, in my humble tests something inferior in the link to the 22/5/17 stock, but this is just a detail. Please go ahead with your great work, and I hope you can continue to improve future versions. Thank you.