Deep Pink

From Chessprogramming wiki
Revision as of 15:20, 31 March 2020 by GerdIsenberg (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Home * Engines * Deep Pink

Deep Pink, [2]
an experimental open source chess engine by Erik Bernhardsson, written in Python as an attempt to learn and play chess. Deep Pink applies negamax alpha-beta and a deep neural network as evaluation function, using Theano, python-chess, and Sunfish [3]. The input representation seems similar to Octavius with 12x64 nodes, not feeding in side to move, castling rights, and en passant target square. The first hidden layer has 2048 neurons as well. Bernhardsson used a GPU instance to train the net with 100M games for about four days using stochastic gradient descent with Nesterov momentum [4] [5] [6] [7].

See also



As the author of sunfish.... by Thomas Ahle, Hacker News, December 2, 2014
This has been tried many times before, with better-but-still-lackluster results.... by halfcat, Hacker News, December 2, 2014

2015 ...

External Links


Up one level