Releases · vshcherbyna/igel

What's new:

extend by 2 in singular extension on non pv move
improve low bound condition in singular extension
check beta score in probcut condition
store in TT in standing pat in qsearch
implement iid
tune null move depth
tune razor depth
remove instant moves handling

Regression run against Igel 3.5.0:

Long Time Control With Increment (60s+0.6s)

Elo   | 21.28 +- 3.60 (95%)
Conf  | 60.0+0.60s Threads=1 Hash=64MB
Games | N: 8860 W: 2461 L: 1919 D: 4480
Penta | [7, 798, 2299, 1298, 28]

Short Time Control With Increment (10s+0.1s)

Elo   | 20.07 +- 3.21 (95%)
Conf  | 10.0+0.10s Threads=1 Hash=8MB
Games | N: 13918 W: 3953 L: 3150 D: 6815
Penta | [76, 1392, 3306, 2023, 162]

What's new:

new network architecture
new network c049c117

Regression run against Igel 3.4.0:

Long Time Control With Increment (60s+0.6s)

ELO   | 43.10 +- 3.70 (95%)
CONF  | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 17568 W: 5631 L: 3463 D: 8474

Short Time Control With Increment (10s+0.1s)

ELO   | 38.72 +- 3.35 (95%)
CONF  | 10.0+0.10s Threads=1 Hash=8MB
GAMES | N: 22976 W: 7639 L: 5089 D: 10248

What's new:

fixing major bug in accumulator handling for null moves
avoid heap objects (std::vector) in search function to reduce heap fragementation
switching to pure nnue evaluation
new network architecture HalfKav2_hm
new network ign-3-debc71a4

Regression run against Igel 3.3.0:

Long Time Control With Increment

ELO   | 75.10 +- 5.38 (95%)
CONF  | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 8480 W: 3133 L: 1328 D: 4019

What's new:

new network architecture
new network ign-2-51ba2968

Regression run against Igel 3.2.0:

Short Time Control With Increment

Score of Igel 3.3.0 64 BMI2 AVX2 vs Igel 3.2.0 64 BMI2 AVX2: 1713 - 756 - 2531  [0.596] 5000
...      Igel 3.3.0 64 BMI2 AVX2 playing White: 1073 - 258 - 1169  [0.663] 2500
...      Igel 3.3.0 64 BMI2 AVX2 playing Black: 640 - 498 - 1362  [0.528] 2500
...      White vs Black: 1571 - 898 - 2531  [0.567] 5000
Elo difference: 67.3 +/- 6.8, LOS: 100.0 %, DrawRatio: 50.6 %

What's new:

smaller margin for prunning quiets
scale down eval to 1000
history extensions only on non pv moves
adjust scale formula
less reductions in lmp when not improving
negative extensions
futility pruning improvements
implement cutnode
new network ign-1-9a48854b created by using prev network (ign-1-3e998e78) and re-trained with a different lambda

Regression run against Igel 3.1.0:

Long Time Control With Increment

ELO   | 43.53 +- 3.13 (95%)
CONF  | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 25000 W: 8152 L: 5036 D: 11812

Short Time Control With Increment

ELO   | 41.22 +- 3.25 (95%)
CONF  | 10.0+0.10s Threads=1 Hash=8MB
GAMES | N: 25008 W: 8572 L: 5619 D: 10817

What's new:

Scale eval at 1280
More aggressive pruning of quiets when not improving
Use 150 cp when razoring
Avoid tt cutoff for rule50
Improve statistics in SMP mode
New network trained using 16B of d5-d8 data from Igel HCE/ign-1-139b702b with d16 validation using ign-1-139b702b

Regression run against Igel 3.0.5:

Long Time Control With Increment

ELO   | 30.13 +- 2.64 (95%)
CONF  | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 20000 W: 3906 L: 2176 D: 13918

Short Time Control With Increment

ELO   | 28.08 +- 3.18 (95%)
CONF  | 10.0+0.10s Threads=1 Hash=8MB
GAMES | N: 20000 W: 5193 L: 3580 D: 11227

This version of Igel won League 1 tournament in TCEC Season 21 - https://en.wikipedia.org/wiki/TCEC_Season_21#League_1

What's new:

Train a new network using Igel 2.6.0 (HCE), Igel 2.9.0 (NNUE) and Igel 3.0.0 (NNUE) data: ign-1-d593efbd
Better time management in sd time control
Skip tt cutoff for null moves in search
Skip null move prunning on tt hit
Vote for best thread in SMP mode
Less aggressive time use in non repeated tc

Regression run against Igel 3.0.0:

Long Time Control No Increment

ELO   | 51.95 +- 2.85 (95%)
SPRT  | 60.0+0.0s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 20548 W: 5296 L: 2246 D: 13006

Long Time Control With Increment

ELO   | 40.20 +- 2.74 (95%)
SPRT  | 60.0+0.6s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 20408 W: 4600 L: 2249 D: 13559

What's new:

Train a new network using Igel 2.6.0 (HCE) and Igel 2.9.0 (NNUE) data: ign-1-139b702b
Remove 'EvalFile' option - as of Igel 3.0.0 the network file is always embedded in the executable
Fix in check extensions
Use incbin on Linux to inject network file into the binary
Tune null move prunning conditions
Tune probcut prunning conditions
Tune quiets prunning
Fix bogus condition for history calculation
Implement more aggressive null move prunning
Stop extension of captures
Prevent explosion of history extensions
Implement better time management for sudden death time control
Count not played quiets for LMP prunning
Remove dummy 100 multiplier for history ordering calculation
Fix compilation issues with gcc 10 and Android NDK's Clang
Remove unused counter move table
Fix out of boundary access for history array
Fix issue with singular extensions - reported by ChizhovVadim (author of CounterGo chess engine)
Fix crash when 'ucinewgame' command is not issued - reported by Ipmanchess
Fix crash when thread position is not initialized - reported by Ipmanchess

Regression run against Igel 2.9.0:

Short Time Control No Increment

ELO   | 115.05 +- 3.88 (95%)
SPRT  | 10.0+0.0s Threads=1 Hash=8MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 20652 W: 9913 L: 3314 D: 7425

Short Time Control With Increment

ELO   | 75.88 +- 3.53 (95%)
SPRT  | 10.0+0.1s Threads=1 Hash=8MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 20904 W: 8061 L: 3567 D: 9276

Long Time Control No Increment

ELO   | 83.16 +- 4.04 (95%)
SPRT  | 60.0+0.0s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 13508 W: 4814 L: 1641 D: 7053

Long Time Control With Increment

ELO   | 58.22 +- 3.81 (95%)
SPRT  | 60.0+0.6s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 13540 W: 4023 L: 1775 D: 7742

What's new:

Switch to own network based on Igel search and evaluation (ign-0-9b1937cc)
NNUE dynamic eval scaling

Regression run against Igel 2.8.0 at 60s+0.6:

ELO   | 23.78 +- 3.56 (95%)
SPRT  | 60.0+0.6s Threads=1 Hash=64MB
LLR   | 0.00 (-2.94, 2.94) [0.00, 0.00]
Games | N: 16112 W: 4102 L: 3001 D: 9009

Regression run against top engines at tc=all/20s+0.2:

Rank Name                          Elo     +/-   Games   Score   Draws
   0 Igel 2.9.0 64 BMI2 AVX2       -84       7    6000   38.2%   40.8%
   1 Stockfish 12 64-bit           387      25    1000   90.3%   18.3%
   2 Komodo 14 64-bit              122      16    1000   66.9%   46.4%
   3 Ethereal 12.75 64-bit          83      16    1000   61.8%   46.9%
   4 Fire 7.1 64-bit                21      16    1000   53.0%   46.3%
   5 Xiphos 0.6 64-bit               5      16    1000   50.7%   45.0%
   6 Komodo 14 MCTS 64-bit         -13      16    1000   48.2%   41.6%

What's new:

Fix null move pruning bug (54 elo)
Use 250cp as lazy eval threshold
Improve time management for sudden death time controls
Use Dietrich Kappe's Night Nurse improved net (internal version ninu-0.2.5b.nnue)

Regression run against Igel 2.7.0 at tc=10+0.1:

Score of Igel 2.8.0 64 BMI2 AVX2 vs Igel 2.7.0 64 BMI2 AVX2: 1012 - 411 - 1380  [0.607] 2803
Elo difference: 75.67 +/- 9.15

Regression run against top engines in tc=all/60+0.6:

Rank Name                          Elo     +/-   Games   Score   Draws
   0 Igel 2.8.0 64 BMI2 AVX2        53       6    8000   57.6%   40.7%
   1 Stockfish 11 64-bit           258      26     500   81.5%   33.0%
   2 Komodo 14 64-bit              110      21     500   65.3%   51.8%
   3 Ethereal 12.25 64-bit          36      22     500   55.2%   48.4%
   4 Komodo 14 MCTS 64-bit          15      22     500   52.2%   48.4%
   5 Xiphos 0.6 64-bit               1      22     500   50.1%   48.6%
   6 Fire 7.1 64-bit               -14      22     500   48.0%   47.6%
   7 RofChade 2.202 64-bit         -63      22     500   41.1%   48.2%
   8 Booot 6.4 64-bit              -76      23     500   39.3%   41.8%
   9 Defenchess 2.2 64-bit         -87      24     500   37.7%   39.8%
  10 Laser 1.7 64-bit              -94      24     500   36.8%   41.2%
  11 Fizbo 2 64-bit               -118      24     500   33.7%   41.4%
  12 RubiChess 1.7.3 64-bit       -122      23     500   33.1%   42.6%
  13 Andscacs 0.95 64-bit         -130      25     500   32.1%   35.8%
  14 Gull 3 64-bit                -151      26     500   29.6%   32.8%
  15 Pedone 2.0 64-bit            -186      27     500   25.5%   30.6%
  16 Strelka 5.5 64-bit           -266      33     500   17.8%   18.8%

Regression run against top engines in tc=all/20+0.2:

Rank Name                          Elo     +/-   Games   Score   Draws
   0 Igel 2.8.0 64 BMI2 AVX2      -179       7    6000   26.3%   38.1%
   1 Stockfish 12 64-bit           458      30    1000   93.3%   13.0%
   2 Komodo 14 64-bit              208      17    1000   76.8%   39.7%
   3 Ethereal 12.75 64-bit         173      16    1000   73.0%   42.7%
   4 Fire 7.1 64-bit               130      17    1000   67.8%   41.1%
   5 Xiphos 0.6 64-bit             121      16    1000   66.7%   47.6%
   6 Komodo 14 MCTS 64-bit         103      16    1000   64.3%   44.3%

Releases: vshcherbyna/igel

Igel 3.6.0

Uh oh!

Igel 3.5.0

Uh oh!

Igel 3.4.0

Uh oh!

Igel 3.3.0

Uh oh!

Igel 3.2.0

Uh oh!

Igel 3.1.0

Uh oh!

Igel 3.0.5

Uh oh!

Igel 3.0.0

Uh oh!

Igel 2.9.0

Uh oh!

Igel 2.8.0

Uh oh!