Interesting game on TCEC

[Dann Corbit]

Alliestein verses Ethereal.
There was a huge jump in the eval of Alliestein and it is taking a long time for the other engines to catch up (Ethereal and the two kibitzer engines).
I wonder what Alliestein saw.

[jorose]

Are you sure its not just the CP conversion formula? I'm not up to date, but last I checked Allie was still using the old LC0 formula, which tended to create rather inflated eval scores.

This is shaping up to be a great DivP, starting with 3 consecutive decisive games.

[Dann Corbit]

On ply 28, I actually heard Allie say, "There's something I ought to tell you... I'm not left handed either!"

[Stephen Ham]

The Princess Bride! Well done, Dann.

Jonathan, the first four games were decisive as I watched game 4. It's possible the 5th was too, but I went to bed and missed it.

Regards,
-Steve-

[Dann Corbit]

The Princess Bride! Well done, Dann.

Jonathan, the first four games were decisive as I watched game 4. It's possible the 5th was too, but I went to bed and missed it.

Regards,
-Steve-

And (incredibly) Stoofvlees won his first two games, starting as black.

[jjoshua2]

Allie eval is re-tuned to match SF between 0 and 4, while still matching its old eval at 10.0. So somewhere in the 6-8 range eval starts going up pretty fast it seems. Max eval now is 256.

[jorose]

Allie eval is re-tuned to match SF between 0 and 4, while still matching its old eval at 10.0. So somewhere in the 6-8 range eval starts going up pretty fast it seems. Max eval now is 256.

Cool, so we will be able to compare nicely until +4 and after that its just kind of like "We are basically just completely winning". Which is pretty much what any score above +4 in TCEC means anyways.

[Dann Corbit]

Allie eval is re-tuned to match SF between 0 and 4, while still matching its old eval at 10.0. So somewhere in the 6-8 range eval starts going up pretty fast it seems. Max eval now is 256.

Cool, so we will be able to compare nicely until +4 and after that its just kind of like "We are basically just completely winning". Which is pretty much what any score above +4 in TCEC means anyways.

I did a curve fit to calculate a centipawn score from wins, losses and draws.
At around +/- 440 centipawns, it goes asymptotic (at least with the Levenberg Marquardt curve fit I used) so the odds of winning after being behind by 440 centipawns is very, very low. I think if I had used floating point math throughout the calculation, I might have gotten a better curve, though.

[jjoshua2]

There was a huge jump in the eval of Alliestein

SF gets over +3 with Ne2 with some depth, actually a bit more than alliestein. The real boom was move 29 though.

Code: Select all

2r1k1nr/nq2bpp1/4p3/1p1pP1Bp/1P1P2P1/2NQ1N1P/5PK1/R6R w k - 0 1
Analysis by Stockfish 260520 64 POPCNT:

28.Ne2 Bxg5 29.Nxg5 Ne7 30.Nf4 Rc4 31.Nxh5 Rxh5 32.gxh5 Nf5 33.Nf3 Nc6 34.Qa3 Rxb4 35.Rhb1 Rxb1 36.Rxb1 b4 37.Qa4 Kd7 38.Rc1 Nfe7 39.Ne1 Qa7 40.Qxa7+ Nxa7 41.Rb1 Nac6 42.Nc2 Nf5 43.Nxb4 Ncxd4 44.Ra1 Ke7 45.Ra7+ Kf8 46.Ra8+ Ke7 47.Nd3 Nc6 48.Rc8 Kd7 49.Rg8 Nce7 50.Rf8 Nh6 51.Ra8 Nc6 52.Kf3 Nd4+ 53.Ke3 Nhf5+ 54.Kf4 Kc6 55.Rb8 Nh6 56.Rb1 Ne2+ 57.Ke3 Nc3 58.Rb8 Ne4 59.Kf4 Nd2 60.Rf8 Kb6 61.Rh8 Kc6 62.Rb8 Nc4 63.Ra8 
  White has a decisive advantage: +- (3.12)  Depth: 51/71   00:16:02  18549MN, tb=6870222

[jjoshua2]

A few minute on that boom and SF is pretty close to the +4 threshold. I think Allie did evaluate a little better enough to get something more like a +5 or 6 and the eval takes off there. Still first to see it was actually won I guess, but it's hard to compare. This is certainly higher eval than crystal got as kibitzer.

Code: Select all

New game Line
Analysis by Stockfish 260520 64 POPCNT:

29.Nf4 Bxg5 30.Nxg5 Ne7 31.Nxh5 Rxh5 32.gxh5 Nf5 33.Nf3 Nc6 34.Qa3 Rxb4 35.Rhb1 Rxb1 36.Rxb1 b4 37.Qa4 Kd7 38.Rc1 Qb6 39.Qa8 b3 40.Qa4 b2 41.Rb1 Nfxd4 42.Nxd4 Qxd4 43.Qxd4 Nxd4 44.Rxb2 Kc6 45.Kg3 Kc7 46.Kf4 Kc6 47.Rb8 Kd7 48.Rb7+ Ke8 49.Rb1 Nf5 50.Rb8+ Kd7 51.Kf3 Nh6 52.Rb6 Ke8 53.Kf4 Nf5 54.Kg5 Ke7 55.Rb7+ Ke8 56.Ra7 Kf8 57.Ra8+ Ke7 58.Kf4 Kd7 59.Ra7+ Ke8 60.Rc7 Nd4 61.Ke3 Nf5+ 62.Kd3 Nh6 63.Kd4 Nf5+ 64.Kc3 
  White is winning: +- (3.71)  Depth: 43/71   00:11:23  8602MN, tb=1205389
(Science, Baltimore 28.05.2020)