Changes

Jump to: navigation, search

Stockfish NNUE

4,322 bytes added, 18:16, 3 August 2021
→‎HalfKA: Sharpening the image
<ref>[http://www.talkchess.com/forum3/viewtopic.php?f=7&t=74531&start=1 Re: NNUE accessible explanation] by [[Jonathan Rosenthal]], [[CCC]], July 23, 2020</ref>.
==HalfKP==
The so called '''HalfKP''' structure consists of two halves covering input layer and first hidden layer, each half of the input layer associated to one of the two [[King|kings]], cross coupled with the side to move or not to move halves of the first hidden layer.
For each either black or white king placement, the 10 none king pieces on their particular squares are the boolean {0,1} inputs, along with a relict from Shogi piece drop (BONA_PIECE_ZERO),
The output of the output layer is divided by FV_SCALE = 16 to produce the NNUE evaluation. SF's evaluation then take some further steps such as adding a Tempo bonus (even though the NNUE evaluation inherently already takes into account the side to move in the "transform" step) and scaling the evaluation towards zero as rule50_count() approaches 50 moves.
 
==HalfKA==
In subsequent Stockfish versions the network architecture was further improved by [[Tomasz Sobczyk]] et al.. The '''HalfKA''' architecture uses 12x64x64 = 45056 inputs for each of the 12 piece types times 64 squares for each of the 64 own king squares, times two, for both the side to move and other side perspective, further using the [[Vertical Flipping|vertical flip]] instead of the HalfKP rotate.
'''HalfKAv2''' as applied in Stockfish '''14''' saves some space considering the king square redundancy using 11x64x64 = 45056 inputs per side, mapped to a 2x520 linear feature transformer <ref>[https://github.aom/official-stockfish/Stockfish/commit/e8d64af1230fdac65bb0da246df3e7abe82e0838 New NNUE architecture and net · official-stockfish/Stockfish@e8d64af · GitHub]</ref> <ref>[https://github.com/official-stockfish/Stockfish/pull/3474 Update default net to nn-8a08400ed089.nnue by Sopel97 · Pull Request #3474 · official-stockfish/Stockfish · GitHub]</ref> <ref>[https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md#halfkav2-feature-set HalfKAv2 feature set | nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub]</ref>, further feeding 8x2 outputs of this feature transformer directly to the output for better learning of unbalanced material configurations <ref>[https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md#a-part-of-the-feature-transformer-directly-forwarded-to-the-output A part of the feature transformer directly forwarded to the output | nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub]</ref>. Another improvement was using eight 512x2->16->32->1 output sub-networks discriminated by (piece_count-1) div 4 in the 0 to 7 range
<ref>[https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md#multiple-psqt-outputs-and-multiple-subnetworks Multiple PSQT outputs and multiple subnetworks | nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub]</ref>.
 
[[FILE:HalfKAv2.png|none|border|text-bottom|658px]]
HalfKAv2 architecture by [[Tomasz Sobczyk]] <ref>[https://user-images.githubusercontent.com/8037982/118656988-553a1700-b7eb-11eb-82ef-56a11cbebbf2.png HalfKAv2.png] Image courtesy by [[Tomasz Sobczyk]]</ref>
=Network=
All networks are Networks were built by some volunteers but not by any big community (differs from [[Leela Chess Zero]]) and can be , uploaded into [[Stockfish#Fishtest|Fishtest]] for testing. Networks with good test results will be are released officially on the Fishtest website <ref>[https://tests.stockfishchess.org/nns Neural Net download and statistics]</ref>with average speed of 2 weeks per network <ref>[https://github.com/official-stockfish/Stockfish/discussions/3628#discussioncomment-1047323 One year of NNUE.... · official-stockfish/Stockfish · GitHub] by [[Joost VandeVondele]], July 26, 2021</ref>. After long discussing the best way to publish networks with Stockfish <ref>[https://github.com/official-stockfish/Stockfish/issues/3030 Improve dealing with the default net? Issue ##3030 · official-stockfish/Stockfish · GitHub] by [[Joost VandeVondele]], August 19, 2020</ref>, the developing team decided to embed the default network into Stockfish binaries, making sure NNUE always works as well as bringing more convenience to users.
After long discussing In late 2020, [[Gary Linscott]] started an implementation of the best way to publish networks with Stockfish NNUE training in [https://en.wikipedia.org/wiki/PyTorch PyTorch] <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=7&t=75724 Pytorch NNUE training] by [[Gary Linscott]], [[CCC]], November 08, 2020</ref> <ref>[https://github.com/officialglinscott/nnue-pytorch GitHub -stockfishglinscott/Stockfishnnue-pytorch: NNUE (Chess evaluation) trainer in Pytorch]</issues/3030 Improve dealing ref> using [[GPU]] resources to efficiently train networks. Further, the collaboration with the default net? Issue ##3030 · official[[Leela Chess Zero]] team in February 2021 <ref>[https://groups.google.com/g/fishcooking/c/AzYDbbv-stockfish/Coo Stockfish · GitHub13] by [[Joost VandeVondele]], August [[Computer Chess Forums|FishCooking]], February 19, 20202021 </ref>payed off, the developing team decided in providing billions of positions to embed train the default network into new networks <ref>[https://stockfishchess.org/blog/2021/stockfish-14/ Stockfish 14], The Stockfish binariesTeam, July 02, making sure NNUE always works as well as bringing more convenience to users2021</ref>.
=Hybrid=
Being attracted by new advantages as well as being encouraged by some impressive successes, many developers joined or continued to work. The [[#Source|Official Stockfish]] repository shows the numbers of commits, ideas increased significantly after merging NNUE.
 
=Elo gain=
[[Joost VandeVondele]] has created a graph to show how Stockfish gains Elo with NNUE after a year:<ref>[https://github.com/official-stockfish/Stockfish/discussions/3628#discussioncomment-1047728 One year of NNUE.... · official-stockfish/Stockfish · GitHub] by [[Joost VandeVondele]], July 26, 2021</ref>
[[FILE:NNEUOneYearEloGain.png|none|border|text-bottom|1024px]]
=Suggestions=
* [http://www.talkchess.com/forum3/viewtopic.php?f=2&t=76833 NNUE Research Project] by [[Ed Schroder|Ed Schröder]], [[CCC]], March 10, 2021 » [[NNUE]]
* [http://www.talkchess.com/forum3/viewtopic.php?f=2&t=76844 NNUE ranking] by Jim Logan, [[CCC]], March 12, 2021
* [http://www.talkchess.com/forum3/viewtopic.php?f=2&t=77344 Stockfish with new NNUE architecture and bigger net released] by [[Stefan Pohl]], [[CCC]], May 19, 2021 <ref>[https://github.com/official-stockfish/Stockfish/pull/3474 Update default net to nn-8a08400ed089.nnue by Sopel97 · Pull Request #3474 · official-stockfish/Stockfish · GitHub] by [[Tomasz Sobczyk]]</ref>
=External Links=
* [https://hxim.github.io/Stockfish-Evaluation-Guide/?p=nnue Stockfish Evaluation Guide] <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=74681 You can now look inside NNUE and look at its Per square value estimation] by [[Henk Drost]], [[CCC]], August 04, 2020</ref>
* [https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md NNUE Guide (nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub)] hosted by [[Gary Linscott]]
* [https://github.com/official-stockfish/Stockfish/discussions/3628 One year of NNUE.... · official-stockfish/Stockfish · GitHub] by [[Joost VandeVondele]], July 26, 2021
 
==Source==
* [https://github.com/official-stockfish/Stockfish GitHub - Official-stockfish]
==Rating==
* [https://github.com/glinscott/fishtest/wiki/Regression-Tests Regression Tests]
* [http://ccrl.chessdom.com/ccrl/404/cgi/engine_details.cgi?match_length=30&each_game=1&print=Details&each_game=1&eng=Stockfish%2014%2064-bit%208CPU#Stockfish_14_64-bit_8CPU Stockfish 14 64-bit 8CPU] in [[CCRL|CCRL Blitz]]* [http://ccrl.chessdom.com/ccrl/404/cgi/engine_details.cgi?print=Details&each_game=1&eng=Stockfish%2012%2064-bit#Stockfish_12_64-bit Stockfish 12 64-bit] in [[CCRL|CCRL Blitz]]
* [http://computerchess.org.uk/ccrl/404/cgi/engine_details.cgi?print=Details&each_game=1&eng=Stockfish%2BNNUE%20150720%2064-bit%204CPU#Stockfish%2BNNUE_150720_64-bit_4CPU Stockfish+NNUE 150720 64-bit 4CPU] in [[CCRL|CCRL Blitz]]
==Misc==
* [https://en.wikipedia.org/wiki/Stockfish Stockfish from Wikipedia]
* [https://en.wikipedia.org/wiki/Nue Nue from Wikipedia]
* [[:Category:Hiromi UeharaSenri Kawaguchi|HiromiSenri Kawaguchi]] - [https://en.wikipediaameliaray.org/wikinet/Spectrum_(Hiromi_album) Spectrumquarantuned The Quarantuned Music Festival], 2019May 2020, [https://en.wikipedia.org/wiki/YouTube YouTube] Video: {{#evu:https://www.youtube.com/watch?v=A8RCz_RoefMjO8VlSqw7do|alignment=left|valignment=top}}
=References=
[[Category:Mac]]
[[Category:Fish]]
[[Category:Hiromi UeharaSenri Kawaguchi]]
Rom77
63
edits
Retrieved from "https://www.chessprogramming.org/Special:MobileDiff/24609...25711"

Navigation menu