That is so awesome - keep up the great work!xr_a_y wrote: ↑Fri May 24, 2019 6:39 am Minic is now better than fruit2.1
Code: Select all
Score of fruit_21 vs Minic: 33 - 65 - 34 [0.379] Elo difference: -85.94 +/- 52.46
A complete 2000 lines of code engine
Moderators: hgm, Rebel, chrisw
-
- Posts: 4889
- Joined: Thu Mar 09, 2006 6:34 am
- Location: Pen Argyl, Pennsylvania
Re: A complete 2000 lines of code engine
-
- Posts: 327
- Joined: Sat Mar 27, 2010 7:15 pm
Re: A complete 2000 lines of code engine
Aha in qsearch, I do eval hashtable store and probe, and main hashtable probe only. But it seems many engines have now got rid of the eval hashtable, and store eval in the main hashtable.xr_a_y wrote: ↑Fri May 24, 2019 9:51 am Using TT in qsearch was useless in Minic with a very simple eval. It became good to store and probe in qsearch when eval started to be the bottleneck.
Just probing TT in qsearch seems strange to me, this will only be good when in check I guess... I store things in qsearch with depth=-1. I guess the important thing is to get eval score for free, I don't expect a lot from TT cut-off.
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
Yes I was using two tables before, but switch to the xiphos style eval+score TT some weeks ago.silentshark wrote: ↑Fri May 24, 2019 4:50 pmAha in qsearch, I do eval hashtable store and probe, and main hashtable probe only. But it seems many engines have now got rid of the eval hashtable, and store eval in the main hashtable.xr_a_y wrote: ↑Fri May 24, 2019 9:51 am Using TT in qsearch was useless in Minic with a very simple eval. It became good to store and probe in qsearch when eval started to be the bottleneck.
Just probing TT in qsearch seems strange to me, this will only be good when in check I guess... I store things in qsearch with depth=-1. I guess the important thing is to get eval score for free, I don't expect a lot from TT cut-off.
-
- Posts: 327
- Joined: Sat Mar 27, 2010 7:15 pm
Re: A complete 2000 lines of code engine
Maybe I should look at this again - is the single table a clear win?xr_a_y wrote: ↑Fri May 24, 2019 5:17 pmYes I was using two tables before, but switch to the xiphos style eval+score TT some weeks ago.silentshark wrote: ↑Fri May 24, 2019 4:50 pmAha in qsearch, I do eval hashtable store and probe, and main hashtable probe only. But it seems many engines have now got rid of the eval hashtable, and store eval in the main hashtable.xr_a_y wrote: ↑Fri May 24, 2019 9:51 am Using TT in qsearch was useless in Minic with a very simple eval. It became good to store and probe in qsearch when eval started to be the bottleneck.
Just probing TT in qsearch seems strange to me, this will only be good when in check I guess... I store things in qsearch with depth=-1. I guess the important thing is to get eval score for free, I don't expect a lot from TT cut-off.
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
not for Minic, it was equal with a poor eval at least.
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
A totally absurd question rise up yesterday : "can fairy-max still beat Minic?". And the answer is YES ! (but not often ...)
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
Be careful testers, version 0.61 of Minic is affected by a bug. May segfault in pawn endgame. I hope it won't affect the current div6 or group C.
A patch will be released soon, 0.60 can be used instead.
Best regards
A patch will be released soon, 0.60 can be used instead.
Best regards
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
Wow crazy bug ... material table was not taking into account the removed pawn after a promotion ... Let's see how this affect elo ...
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
Some SMP results :
Code: Select all
Rank Name Elo +/- Games Score Draws
1 Minic 3 threads 77 23 598 61.0% 35.3%
2 rofChade1 71 25 598 60.1% 22.2%
3 Minic 2 threads 46 23 599 56.5% 34.6%
4 Topple_0.4.0_nehalem 5 25 598 50.7% 21.4%
5 Minic -88 23 598 37.5% 32.9%
6 fruit_21 -113 26 599 34.2% 19.7%
1795 of 15000 games finished.
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
Re: A complete 2000 lines of code engine
Release 0.62 is available and fix this bug.
Please note that GroupeC and Division6 can of course go on with 0.61 but after that 0.61 shall not be tested more.