TalkChess.com

> One typo in line #3, one spurious space : "arasan 21.3" -> "arasan21.3"
Fixed.

Windows Defender is bundled with Windows now. That is probably all you need.

From a recent match Arasan 21.2 - Komodo Dragon: 2r2rk1/pb2qppp/4p3/1p6/1PB5/P1R1PN2/6PP/3R1QK1 w - - Black has just offered a pawn with .. b5. If White takes, he can get two minors for a Rook after 1. Bxb5 Rxc3 2. Rd7 Qe8 3. Rxb7. Arasan took the pawn (Bxb5) and lost. The longer-term consequences o...

One typo in line #3, one spurious space : "arasan 21.3" -> "arasan21.3"

Fixed, thanks.

I am also going to suggest replacements for numbers 65 and 71: 8/8/Q4p1k/6p1/4pNqn/PP2P1P1/3r1P2/2R3K1 b - - bm gxf4; id "arasan20.65"; c0 "Arasan-Wasp, FICS 2020 (analysis)"; 5k2/2N2p2/1p1p1nr1/p2P4/P1P3Pb/1P2Q3/1q6/3B1RK1 b - - bm b5; id "arasan20.71"; c0 "Ethereal 11.0-Stockfish 9, Andreas St\ ra...

I would like to suggest the following replacements for the problematic positions. These are from a list of candidate test positions I have been collecting. None are super-hard but difficult enough to be interesting IMO. r2r1k2/p3qpp1/1p1ppn1p/n5B1/P1PNbP2/2P3Q1/4B1PP/4RRK1 w - - bm Bg4; id "arasan 2...

Texel tuning is basically supervised learning using logistic regression. There is a very large literature on this, outside of the field of chess.

For just one example: https://web.stanford.edu/~jurafsky/slp3/5.pdf

Another couple positions that were mentioned: r1q1k2r/1p1nbpp1/2p2np1/p1Pp4/3Pp3/P1N1P1P1/1P1B1P1P/R2QRBK1 b kq - bm Bxc5; id "arasan20.3"; c0 "Mapanare-XXAAXX, Tues 23 Feb Rapid Engine Tour, Infinity Chess 2016"; depth 50 seldepth 13 multipv 1 score +213 nodes 32640209729 nps 18133439 hashfull 1000...

Erroneous positions I found : Multi-PV analysis of these with Stockfish-12, dual Xeon e5-2690v3 (16 cores, 16G hash), 1/2 hour per position: engine ready 5rk1/8/pqPp1r1p/1p1Pp1bR/4B3/5PP1/PP2Q1K1/R7 w - - bm Rxg5+; id "arasan20.61"; c0 "sedarpi-noamgmi, www.chesspresso.net 2014"; depth 48 seldepth ...

Gradient descent can be run on parallel threads if you are using an algorithm like ADAM. Does not readily work with "mini-batches," though.

--Jon