Match between SF 10 and Fat Fritz

[Steppenwolf]

https://en.chessbase.com/post/fat-fritz-and-the-kid

[Steppenwolf]

Interesting to me is the different evaluation of the two positions.
Why is SF so wrong? Horizon Effect?

[zullil]

Interesting to me is the different evaluation of the two positions.
Why is SF so wrong? Horizon Effect?

Human error for the first, at least:

Sf-dev says:

info depth 44 seldepth 81 multipv 1 score cp -164 nodes 10985542559 nps 20362904 hashfull 1000 tbhits 0 time 539488 pv b7c6 d5c6 d7b6 a2a4 c8e6 a4a5 b6c4 d1b3 d6d5 a1d1 c4e3 e4d5 e6g8 e1d3 f6e7 b3b2 e7d6 c3e4 d8e7 e4d6 c7d6 b4b5 g8d5 d3b4 d5f7 b5b6 e3d1 e2d1 a7b6 a5b6 f8b8 b6b7 b8b7 c6b7 e7b7 b2d2 b7b5 d1e2 b5c5 e2d3 a8a4 f1b1 c5e3 d2e3 f4e3 h1h2 f7g6 d3g6 g7g6 b4d5 a4a3 b1e1 e3e2 e1e2 a3f3 e2e3 f3f2 h2h3 g6g5 h3g3 f2f1 e3e4 f1g1 g3f3

So +1.64 for WHITE.

[Vinvin]

Interesting to me is the different evaluation of the two positions.
Why is SF so wrong? Horizon Effect?

Human error for the first, at least:

Sf-dev says:

info depth 44 seldepth 81 multipv 1 score cp -164 nodes 10985542559 nps 20362904 hashfull 1000 tbhits 0 time 539488 pv b7c6 d5c6 d7b6 a2a4 c8e6 a4a5 b6c4 d1b3 d6d5 a1d1 c4e3 e4d5 e6g8 e1d3 f6e7 b3b2 e7d6 c3e4 d8e7 e4d6 c7d6 b4b5 g8d5 d3b4 d5f7 b5b6 e3d1 e2d1 a7b6 a5b6 f8b8 b6b7 b8b7 c6b7 e7b7 b2d2 b7b5 d1e2 b5c5 e2d3 a8a4 f1b1 c5e3 d2e3 f4e3 h1h2 f7g6 d3g6 g7g6 b4d5 a4a3 b1e1 e3e2 e1e2 a3f3 e2e3 f3f2 h2h3 g6g5 h3g3 f2f1 e3e4 f1g1 g3f3

So +1.64 for WHITE.

Position :
[d]r1bq1r2/pppn2k1/2Pp1b2/3Pp3/1P2PpP1/2N2P1p/P3B3/R2QNR1K b - - 0

[zullil]

Interesting to me is the different evaluation of the two positions.
Why is SF so wrong? Horizon Effect?

Human error for the first, at least:

Sf-dev says:

info depth 44 seldepth 81 multipv 1 score cp -164 nodes 10985542559 nps 20362904 hashfull 1000 tbhits 0 time 539488 pv b7c6 d5c6 d7b6 a2a4 c8e6 a4a5 b6c4 d1b3 d6d5 a1d1 c4e3 e4d5 e6g8 e1d3 f6e7 b3b2 e7d6 c3e4 d8e7 e4d6 c7d6 b4b5 g8d5 d3b4 d5f7 b5b6 e3d1 e2d1 a7b6 a5b6 f8b8 b6b7 b8b7 c6b7 e7b7 b2d2 b7b5 d1e2 b5c5 e2d3 a8a4 f1b1 c5e3 d2e3 f4e3 h1h2 f7g6 d3g6 g7g6 b4d5 a4a3 b1e1 e3e2 e1e2 a3f3 e2e3 f3f2 h2h3 g6g5 h3g3 f2f1 e3e4 f1g1 g3f3

So +1.64 for WHITE.

[d]r1bq1r2/pppn2k1/2Pp1b2/3Pp3/1P2PpP1/2N2P1p/P3B3/R2QNR1K b - - 0

I'm searching this position. Am I wrong?

Code: Select all

 +---+---+---+---+---+---+---+---+
 | r |   | b | q |   | r |   |   |
 +---+---+---+---+---+---+---+---+
 | p | p | p | n |   |   | k |   |
 +---+---+---+---+---+---+---+---+
 |   |   | P | p |   | b |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   | P | p |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   | P |   |   | P | p | P |   |
 +---+---+---+---+---+---+---+---+
 |   |   | N |   |   | P |   | p |
 +---+---+---+---+---+---+---+---+
 | P |   |   |   | B |   |   |   |
 +---+---+---+---+---+---+---+---+
 | R |   |   | Q | N | R |   | K |
 +---+---+---+---+---+---+---+---+

Fen: r1bq1r2/pppn2k1/2Pp1b2/3Pp3/1P2PpP1/2N2P1p/P3B3/R2QNR1K b - - 1 20
Key: 09766F38D7B5B8F8

info depth 49 seldepth 81 multipv 1 score cp -189 nodes 27399877950 nps 20439357 hashfull 1000 tbhits 0 time 1340545 pv b7c6 d5c6 d7b6 a2a4 c8e6 a4a5 b6c4 d1b3 d6d5 a1d1 c4e3 e4d5 e6g8 e1d3 f6e7 d3e5 d8d6 e5d7 d6b4 b3b4 e7b4 d7f8 g7f8 c3b5 b4a5 d1d4 e3f1 e2f1 a5b6 d4f4 f8g7 f1c4 a8d8 f4f5 g8f7 f5g5 g7f6 f3f4 a7a6 b5c3 f7g6 g5e5 b6c5 h1h2 c5d6 e5e6 f6g7 c3e2 g6f7 e6e4 f7g6 e4e3 a6a5 h2h3 d6c5 e3e6 d8b8 e2g3 a5a4 g4g5 a4a3 f4f5 g6f7 e6e2 b8b4 c4a2 b4b2 g5g6 b2e2 g3e2


Black has the move, so cp = -189 means WHITE is +1.89.

[Laskos]

So, Chessbase marketing of Fat Fritz started in full. The regular Lc0 branch of T499xx nets does use 100 most popular human 4-plier openings, and if one is curious about KID specifics, main variations of it could have been added as openings. I am not sure Fat Fritz plays KID any better than 499xx nets, or even main run 42xxx nets or JH latest bignet. And overall, Fat Fritz might play worse both the openings and Chess in general than the main run late 42xxx nets. SL hardly got any spectacular result with Leela based NN engines.

[dkappe]

So, Chessbase marketing of Fat Fritz started in full. The regular Lc0 branch of T499xx nets does use 100 most popular human 4-plier openings, and if one is curious about KID specifics, main variations of it could have been added as openings. I am not sure Fat Fritz plays KID any better than 499xx nets, or even main run 42xxx nets or JH latest bignet. And overall, Fat Fritz might play worse both the openings and Chess in general than the main run late 42xxx nets. SL hardly got any spectacular result with Leela based NN engines.

The weakness of SL has more to do with the bad tools than an inherent problem with SL. I’ve developed a tool that allows me to generate V4 data using pgn’s and sf10 multipv. Head to head, the 128x10 CCRL net is 250 elo weaker than Bad Gyal 7 (where the source material is unfiltered lichess games). And mind you, the sf10 multipv is only at depth 11.