MadChess 3.1 Released

[emadsen]

I have released a new version of my engine. MadChess 3.1 is about 100 Elo stronger than the prior release, 3.0. You can read details on the MadChess 3.1 Released post on my blog.

I estimate MadChess plays with the following strength.

• Bullet (2m + 1s) = 2706 Elo
• Blitz (5m + 3s) = 2742 Elo

Enjoy!

I'm glad to see so many new programmers have joined our chess programming family in the last couple years. While I don't always comment, I do read their forum posts with interest. Keep up the good work.

[Graham Banks]

Thanks, Erik.

[AndrewGrant]

Congrats, Erik. Always happy to see your name pop up on talkchess. Blog post(s) and linked commits provided! Hope you are doing well

[lithander]

Great news! Right around the time when Leorik has finally caught up with MadChess a new versions comes in! Perfect timing!

A head-to-head race with an actively developed engine is much more fun to me than passing by stable engines on the CCRL list. Luckily the Division 7 (where both MadChess and Leorik were playing last Season) seemed very active and a lot of the engines there had seen a major strength-increase recently. Very motivating!

[Steve Maughan]

Big congrats Erik — you've made some excellent progress over the last couple of years.

I ran a few test positions. One that MadChess didn't handle well was the classic Lasker-Reichhelm position:

[d]8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - -

MadChess wants to plat Kb2 and is super-convinced it's winning but in reality leads to a draw — Kb1 is the only win. I've used this position to debug hash tables. Of course tuning for one position is a terrible idea, but in this case MadChess is incorrectly seeing a win, which IMHO is a reason to investigate. It might be worthwhile popping the proverbial hood and taking a look.

Congrats again — you're motivating me to dust off Maverick and see what I can do.

Steve

[emadsen]

I ran a few test positions. One that MadChess didn't handle well was the classic Lasker-Reichhelm position:

[d]8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - -

MadChess wants to plat Kb2 and is super-convinced it's winning but in reality leads to a draw — Kb1 is the only win. I've used this position to debug hash tables.
Steve

Thanks all.

Interesting. Thanks, Steve, for testing and sharing. I know MadChess used to solve this position. I don't think it's a hashtable bug. I suspect my endgame evaluation scaling code is to blame. I don't like the code- it creates inflated endgame scores- but it delivers Elo. I need to rewrite that code and fix this issue in the process.

[Tearth]

Glad to see MadChess is getting stronger, your blog was insanely useful when I was learning about chess engines a few years ago. Thank you and good luck with further work!

[lithander]

Code: Select all

tc=40/60 10MB Hash
Score of Leorik-2.2 vs MadChess-3.1: 88 - 67 - 45  [0.552] 200
...      Leorik-2.2 playing White: 42 - 39 - 19  [0.515] 100
...      Leorik-2.2 playing Black: 46 - 28 - 26  [0.590] 100
...      White vs Black: 70 - 85 - 45  [0.463] 200
Elo difference: 36.6 +/- 42.7, LOS: 95.4 %, DrawRatio: 22.5 %

tc=40/20 10MB Hash
Score of Leorik-2.2 vs MadChess-3.1: 210 - 108 - 82  [0.627] 400
...      Leorik-2.2 playing White: 120 - 59 - 21  [0.652] 200
...      Leorik-2.2 playing Black: 90 - 49 - 61  [0.603] 200
...      White vs Black: 169 - 149 - 82  [0.525] 400
Elo difference: 90.6 +/- 31.2, LOS: 100.0 %, DrawRatio: 20.5 %

The PV bug is gone and I saw now crashes, losses on time or any other stability issues!

And it still seems true for this version that it thrives on longer time controls and doesn't like very fast games so much.

[emadsen]

The PV bug is gone and I saw now crashes, losses on time or any other stability issues!

And it still seems true for this version that it thrives on longer time controls and doesn't like very fast games so much.

Hmm. I haven't seen any crashes since I fixed this bug. But I haven't played any games faster than 2m + 1s. What's 40/20? Is that 40 moves in 20 seconds, repeating? If that's true, that's a faster time control than I use. Perhaps there's an issue related to fast or non-increment time control.

Or am I misinterpreting your statement? "The PV bug is gone and I saw NOW crashes." Did you mean to say you "saw NO crashes"?

[lithander]

The PV bug is gone and I saw now crashes, losses on time or any other stability issues!

And it still seems true for this version that it thrives on longer time controls and doesn't like very fast games so much.

Hmm. I haven't seen any crashes since I fixed this bug. But I haven't played any games faster than 2m + 1s. What's 40/20? Is that 40 moves in 20 seconds, repeating? If that's true, that's a faster time control than I use. Perhaps there's an issue related to fast or non-increment time control.

I‘m too dumb to type: saw no crashes! The w was not meant to be there