Difference between revisions of "Allie"

From Chessprogramming wiki
Jump to: navigation, search
Line 6: Line 6:
  
 
=AllieStein=
 
=AllieStein=
AllieStein is the combination of Allie with Leela's third party '''Leelenstein''' network <ref>[https://github.com/LeelaChessZero/lc0/wiki/Third-Party-Nets Third Party Nets · LeelaChessZero/lc0 Wiki · GitHub]</ref> <ref>[https://www.patreon.com/jjosh Jjosh is creating Leelenstein | Patreon]</ref>, which is trained by [[Supervised Learning|supervised learning]] feeding in games from [[CCRL]],  
+
AllieStein is the combination of Allie with Leela's third party '''Leelenstein''' network by [[Mark Jordan]] <ref>[https://github.com/LeelaChessZero/lc0/wiki/Third-Party-Nets Third Party Nets · LeelaChessZero/lc0 Wiki · GitHub]</ref> <ref>[https://www.patreon.com/jjosh Jjosh is creating Leelenstein | Patreon]</ref>, which is trained by [[Supervised Learning|supervised learning]] feeding in games from [[CCRL]],  
 
supported by '''SGDR''' ([https://en.wikipedia.org/wiki/Stochastic_gradient_descent Stochastic Gradient Descent] with Warm Restarts) <ref>[[Ilya Loshchilov]], [[Frank Hutter]] ('''2016'''). ''SGDR: Stochastic Gradient Descent with Warm Restarts''. [https://arxiv.org/abs/1608.03983 arXiv:1608.03983]</ref>
 
supported by '''SGDR''' ([https://en.wikipedia.org/wiki/Stochastic_gradient_descent Stochastic Gradient Descent] with Warm Restarts) <ref>[[Ilya Loshchilov]], [[Frank Hutter]] ('''2016'''). ''SGDR: Stochastic Gradient Descent with Warm Restarts''. [https://arxiv.org/abs/1608.03983 arXiv:1608.03983]</ref>
 
and '''GGT''' (full-matrix adaptive [https://en.wikipedia.org/wiki/Regularization_(mathematics) regularization]) <ref>[[Naman Agarwal]], [[Brian Bullins]], [[Xinyi Chen]], [[Elad Hazan]], [[Karan Singh]], [[Cyril Zhang]], [[Yi Zhang]] ('''2018'''). ''The Case for Full-Matrix Adaptive Regularization''. [https://arxiv.org/abs/1806.02958 arXiv:1806.02958]</ref>,
 
and '''GGT''' (full-matrix adaptive [https://en.wikipedia.org/wiki/Regularization_(mathematics) regularization]) <ref>[[Naman Agarwal]], [[Brian Bullins]], [[Xinyi Chen]], [[Elad Hazan]], [[Karan Singh]], [[Cyril Zhang]], [[Yi Zhang]] ('''2018'''). ''The Case for Full-Matrix Adaptive Regularization''. [https://arxiv.org/abs/1806.02958 arXiv:1806.02958]</ref>,

Revision as of 08:34, 11 September 2019

Home * Engines * Allie

Allie,
an UCI compliant open source chess engine by Adam Treat, written in C++ using Qt, released under the terms of GPL version 3. Allie is inspired by the seminal AlphaZero paper [1] and the Leela Chess Zero project - in particular it probes the same deep convolutional neural networks produced by Leela Chess, and replaces Lc0 with an own implementation of a PUCT Monte-Carlo tree search [2].

AllieStein

AllieStein is the combination of Allie with Leela's third party Leelenstein network by Mark Jordan [3] [4], which is trained by supervised learning feeding in games from CCRL, supported by SGDR (Stochastic Gradient Descent with Warm Restarts) [5] and GGT (full-matrix adaptive regularization) [6], using batch renormalization [7], and adding gradient noise [8].

See also

Forum Posts

External Links

Chess Engine

Misc

References

  1. David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, Timothy Lillicrap, Karen Simonyan, Demis Hassabis (2017). Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm. arXiv:1712.01815
  2. allie/node.h at master · manyoso/allie · GitHub
  3. Third Party Nets · LeelaChessZero/lc0 Wiki · GitHub
  4. Jjosh is creating Leelenstein | Patreon
  5. Ilya Loshchilov, Frank Hutter (2016). SGDR: Stochastic Gradient Descent with Warm Restarts. arXiv:1608.03983
  6. Naman Agarwal, Brian Bullins, Xinyi Chen, Elad Hazan, Karan Singh, Cyril Zhang, Yi Zhang (2018). The Case for Full-Matrix Adaptive Regularization. arXiv:1806.02958
  7. Sergey Ioffe (2017). Batch Renormalization: Towards Reducing Minibatch Dependence in Batch-Normalized Models. arXiv:1702.03275
  8. Arvind Neelakantan, Luke Vilnis, Quoc V. Le, Ilya Sutskever, Lukasz Kaiser, Karol Kurach, James Martens (2015). Adding Gradient Noise Improves Learning for Very Deep Networks. arXiv:1511.06807

Up one level