Pedantic Developer's Log Stardate...

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

adityachandra
Posts: 22
Joined: Sun Apr 23, 2023 9:26 am
Full name: Aditya Chandra

Re: Pedantic Developer's Log Stardate...

Post by adityachandra »

Thanks, I didn't realise I could have just reconstructed the fens once instead of every time when computing the error :mrgreen:
Just running it I get through my data a lot faster now!
Ras
Posts: 2528
Joined: Tue Aug 30, 2016 8:19 pm
Full name: Rasmus Althoff

Re: Pedantic Developer's Log Stardate...

Post by Ras »

There's another trick with Texel tuning: not all parameters are applicable in every position. Actually, most of them aren't in a specific position. Like, the PSQT value for a king on f2 will only matter if the position has a king on f2. If not, it means that the error contribution from this position for this parameter will not change if you modify the associated PSQT value. Or, the value of the bishop pair only matters in positions where exactly one side has the bishop pair.
Rasmus Althoff
https://www.ct800.net
JoAnnP38
Posts: 250
Joined: Mon Aug 26, 2019 4:34 pm
Location: Clearwater, Florida USA
Full name: JoAnn Peeler

Re: Pedantic Developer's Log Stardate...

Post by JoAnnP38 »

A I prepare to release Pedantic 0.4 in the next couple of days, I thought I would share with you my original bullet list of my plan for the Pedantic 0.4 release marked up to reflect what is actually being released. What is it that they say? That no battle plan survives first contact with the enemy? Somehow that always feels relevant when releasing software.

Image

As I complete some final testing and administration needed for every release, I can share a gauntlet that indicates Pedantic 0.4's new strength of 2819 +/-15 Elo.

Code: Select all

TC 120+1
Rank Name               Elo     +/-   Games    Wins  Losses   Draws   Points   Score    Draw
   0 Pedantic 0.4       -15      15    1400     458     517     425    670.5   47.9%   30.4%
   1 Vengeance 2.0      156      65     100      60      18      22     71.0   71.0%   22.0%
   2 Viridithas 4.0     156      59     100      55      13      32     71.0   71.0%   32.0%
   3 Toga II 4.0         96      58     100      48      21      31     63.5   63.5%   31.0%
   4 Protector 1.4       81      61     100      49      26      25     61.5   61.5%   25.0%
   5 Weiawaga 5.0        60      53     100      38      21      41     58.5   58.5%   41.0%
   6 Willow 2.8          35      56     100      38      28      34     55.0   55.0%   34.0%
   7 Protej 0.6.4        28      58     100      39      31      30     54.0   54.0%   30.0%
   8 Counter 3.2         10      58     100      37      34      29     51.5   51.5%   29.0%
   9 Minic 0.92         -17      57     100      32      37      31     47.5   47.5%   31.0%
  10 Leorik 2.4         -49      58     100      28      42      30     43.0   43.0%   30.0%
  11 Amoeba 2.3         -53      55     100      24      39      37     42.5   42.5%   37.0%
  12 Igel 1.8.3         -78      59     100      25      47      28     39.0   39.0%   28.0%
  13 Zangdar 2.17.09    -85      58     100      23      47      30     38.0   38.0%   30.0%
  14 Stash 21.2        -119      62     100      21      54      25     33.5   33.5%   25.0%
  
2834 - 15 = 2819
User avatar
emadsen
Posts: 438
Joined: Thu Apr 26, 2012 1:51 am
Location: Oak Park, IL, USA
Full name: Erik Madsen

Re: Pedantic Developer's Log Stardate...

Post by emadsen »

JoAnnP38 wrote: Thu Aug 03, 2023 5:46 am As I complete some final testing and administration needed for every release, I can share a gauntlet that indicates Pedantic 0.4's new strength of 2819 +/-15 Elo.
Awesome progress!
Erik Madsen | My C# chess engine: https://www.madchess.net
User avatar
lithander
Posts: 881
Joined: Sun Dec 27, 2020 2:40 am
Location: Bremen, Germany
Full name: Thomas Jahn

Re: Pedantic Developer's Log Stardate...

Post by lithander »

Also noteworthy how 0.4 won the Amateur Series Division 8 after it replaced 0.3 mid tournament! :)
Minimal Chess (simple, open source, C#) - Youtube & Github
Leorik (competitive, in active development, C#) - Github & Lichess
JoAnnP38
Posts: 250
Joined: Mon Aug 26, 2019 4:34 pm
Location: Clearwater, Florida USA
Full name: JoAnn Peeler

Re: Pedantic Developer's Log Stardate...

Post by JoAnnP38 »

lithander wrote: Sun Aug 06, 2023 5:47 pm Also noteworthy how 0.4 won the Amateur Series Division 8 after it replaced 0.3 mid tournament! :)
I wish that were true, but after Akimbo was also replaced (0.4.1 for 0.4.0) both Leorik and Coil now share the lead. I still have a chance, but the window is closing.

You can still catch the tournament live as it is still play with over 90+ games to go here.

I am also afraid that my early Elo estimate may not be consistent across different environments as it played worse in Lars testing environment than on mine and a couple of other developers. I would have loved to have dug into it more with Lars, but I feel that he likes keeping his system "proprietary." lol
User avatar
lithander
Posts: 881
Joined: Sun Dec 27, 2020 2:40 am
Location: Bremen, Germany
Full name: Thomas Jahn

Re: Pedantic Developer's Log Stardate...

Post by lithander »

Oh, I mistook Graham's update post for the final score. I had been on vacation and didn't pay proper attention! Well... then... Go! Leorik ;)
Minimal Chess (simple, open source, C#) - Youtube & Github
Leorik (competitive, in active development, C#) - Github & Lichess
JoAnnP38
Posts: 250
Joined: Mon Aug 26, 2019 4:34 pm
Location: Clearwater, Florida USA
Full name: JoAnn Peeler

Re: Pedantic Developer's Log Stardate...

Post by JoAnnP38 »

Preparing Pedantic 0.5.0 for release. Should take 2-4 days to complete all the testing. One unplanned enhancement was adding Lazy SMP support as I was inspired by Leorik's success in this area. Here are some self-play testing results for this new feature. All testing performed on an AMD Ryzen 9 7940hs with 128mb x thread count for hash table size. Comparison testing was against the last development version committed prior to this effort.

Code: Select all

Lazy SMP 1 thread +8 Elo 
Lazy SMP 2 threads +70 Elo vs 1 thread
Lazy SMP 4 threads +110 Elo vs 1 thread
Lazy SMP 8 threads +169 Elo vs 1 thread

Score of Pedantic Dev vs Pedantic Base: 1684 - 1546 - 3170  [0.511] 6400
...      Pedantic Dev playing White: 978 - 636 - 1587  [0.553] 3201
...      Pedantic Dev playing Black: 706 - 910 - 1583  [0.468] 3199
...      White vs Black: 1888 - 1342 - 3170  [0.543] 6400
Elo difference: 7.5 +/- 6.0, LOS: 99.2 %, DrawRatio: 49.5 %
SPRT: llr 2.2 (100.3%), lbound -2.2, ubound 2.2 - H1 was accepted

Score of Pedantic Dev vs Pedantic Base: 257 - 121 - 308  [0.599] 686
...      Pedantic Dev playing White: 143 - 52 - 148  [0.633] 343
...      Pedantic Dev playing Black: 114 - 69 - 160  [0.566] 343
...      White vs Black: 212 - 166 - 308  [0.534] 686
Elo difference: 69.8 +/- 19.4, LOS: 100.0 %, DrawRatio: 44.9 %
SPRT: llr 2.21 (100.5%), lbound -2.2, ubound 2.2 - H1 was accepted

Score of Pedantic Dev vs Pedantic Base: 149 - 37 - 180  [0.653] 366
...      Pedantic Dev playing White: 86 - 14 - 83  [0.697] 183
...      Pedantic Dev playing Black: 63 - 23 - 97  [0.609] 183
...      White vs Black: 109 - 77 - 180  [0.544] 366
Elo difference: 109.8 +/- 25.3, LOS: 100.0 %, DrawRatio: 49.2 %
SPRT: llr 2.21 (100.4%), lbound -2.2, ubound 2.2 - H1 was accepted

Score of Pedantic Dev vs Pedantic Base: 130 - 17 - 103  [0.726] 250
...      Pedantic Dev playing White: 75 - 7 - 43  [0.772] 125
...      Pedantic Dev playing Black: 55 - 10 - 60  [0.680] 125
...      White vs Black: 85 - 62 - 103  [0.546] 250
Elo difference: 169.3 +/- 33.6, LOS: 100.0 %, DrawRatio: 41.2 %
SPRT: llr 2.2 (100.1%), lbound -2.2, ubound 2.2 - H1 was accepted
JoAnnP38
Posts: 250
Joined: Mon Aug 26, 2019 4:34 pm
Location: Clearwater, Florida USA
Full name: JoAnn Peeler

Re: Pedantic Developer's Log Stardate...

Post by JoAnnP38 »

Announcement: Official Release of Pedantic 0.5.0 is Now Available
I am happy to announce that Pedantic 0.5.0 is now available. Pedantic 0.5.0 is a UCI engine that I have tested in both Arena and CuteChess. This release is estimated to be stronger than v0.4.1 by about 162 Elo and playing strength is estimated to be about 2944 at 40/15 and 3007 at blitz. I am looking forward to seeing how close my estimate is to the one CCRL eventually calculates.

Download Pedantic 0.5.0
User avatar
lithander
Posts: 881
Joined: Sun Dec 27, 2020 2:40 am
Location: Bremen, Germany
Full name: Thomas Jahn

Re: Pedantic Developer's Log Stardate...

Post by lithander »

Congrats on the release! That's another huge improvement over the previous version! Reaching 3000 Elo without NNUE is quiet a feat, imo.
Minimal Chess (simple, open source, C#) - Youtube & Github
Leorik (competitive, in active development, C#) - Github & Lichess