Difference between revisions of "Koivisto"
GerdIsenberg (talk | contribs) |
GerdIsenberg (talk | contribs) |
||
| Line 9: | Line 9: | ||
The [[Bitboards|bitboard]] engine provides [[Automated Tuning|automated evaluation tuning]] by [[Automated Tuning#LogisticRegression|logistic regression]], | The [[Bitboards|bitboard]] engine provides [[Automated Tuning|automated evaluation tuning]] by [[Automated Tuning#LogisticRegression|logistic regression]], | ||
either using [https://en.wikipedia.org/wiki/Stochastic_gradient_descent stochastic gradient descent] or [https://en.wikipedia.org/wiki/Stochastic_gradient_descent#AdaGrad AdaGrad]. | either using [https://en.wikipedia.org/wiki/Stochastic_gradient_descent stochastic gradient descent] or [https://en.wikipedia.org/wiki/Stochastic_gradient_descent#AdaGrad AdaGrad]. | ||
| − | + | Koivisto '''5.0''' in July 2021 replaced the [[Neural Networks#Perceptron|perceptron]] approach of a real-men-evaluation (RME) by an own [[NNUE]] <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=75001&start=4 Re: Koivisto] by [[Finn Eggers]], [[CCC]], September 04, 2020</ref> <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=77664 Koivisto 5.0] by [[Finn Eggers]], [[CCC]], July 07, 2021</ref>. | |
=Features= | =Features= | ||
| Line 43: | Line 43: | ||
** [[Quiescence Search]] | ** [[Quiescence Search]] | ||
==[[Evaluation]]== | ==[[Evaluation]]== | ||
| + | * [[NNUE]] (5.0) | ||
* [[Tapered Eval]] | * [[Tapered Eval]] | ||
* [[Piece-Square Tables]] of [[Float|floats]] | * [[Piece-Square Tables]] of [[Float|floats]] | ||
| Line 97: | Line 98: | ||
[[Category:GPL]] | [[Category:GPL]] | ||
[[Category:UCI]] | [[Category:UCI]] | ||
| + | [[Category:NNUE]] | ||
[[Category:Linux]] | [[Category:Linux]] | ||
[[Category:Windows]] | [[Category:Windows]] | ||
Revision as of 09:31, 8 July 2021
Koivisto,
an UCI compliant open source chess engine by Kim Kåhre and Finn Eggers, at times supported by Eugenio Bruno, written in C++,
and first released on GitHub in September 2020 under the GPL v3.0.
The bitboard engine provides automated evaluation tuning by logistic regression,
either using stochastic gradient descent or AdaGrad.
Koivisto 5.0 in July 2021 replaced the perceptron approach of a real-men-evaluation (RME) by an own NNUE [2] [3].
Contents
Features
Board Representation
Search
- Lazy SMP (2.0)
- Iterative Deepening
- Aspiration Windows (3.0)
- Principal Variation Search
- Transposition Table
- Move Ordering
- Selectivity
Evaluation
- NNUE (5.0)
- Tapered Eval
- Piece-Square Tables of floats
- Mobility
- Tactical Patterns
- Evaluation Patterns
- Pawn Structure
- King Safety
- Tempo (3.0)
- Draw Evaluation (3.0)
- Float SSE Dot products [4]
- Automated Tuning by Logistic Regression
Misc
- Syzygy Bases via Fathom
Forum Posts
- Re: New engine releases 2020 by Finn Eggers, CCC, September 02, 2020
- Re: New engine releases 2020 by Finn Eggers, CCC, September 03, 2020
- Koivisto by Gerd Isenberg, CCC, September 04, 2020
- Re: Koivisto by Finn Eggers, CCC, September 04, 2020
- Koivisto 2.0 by Finn Eggers, CCC, September 08, 2020
- Koivisto 3.0 (Including binaries) by Kim Kåhre, CCC, October 17, 2020
- Koivisto v4.0 release by Eugenio Bruno, CCC, November 09, 2020
- Koivisto 5.0 by Finn Eggers, CCC, July 07, 2021
External Links
Chess Engine
Misc
References
- ↑ Hjalmar Munsterhjelm - Koivikko ja ruispelto (Birch and rye field), 1876, Finnish National Gallery, Wikimedia Commons
- ↑ Re: Koivisto by Finn Eggers, CCC, September 04, 2020
- ↑ Koivisto 5.0 by Finn Eggers, CCC, July 07, 2021
- ↑ Dot products using _mm_mul_ps, _mm_add_ps
