Winning after Qf6+!

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Master Om
Posts: 335
Joined: Wed Nov 24, 2010 9:57 am
Location: INDIA

Re: Winning after Qf6+!

Post by Master Om » Thu Oct 24, 2019 7:26 pm

Always Expect the Unexpected

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Winning after Qf6+!

Post by jp » Thu Oct 24, 2019 7:34 pm

Master Om wrote:
Wed Oct 16, 2019 5:11 pm
Okay. I am using SF Dev to analyse now. Lets see what happens.
Was SF dev the engine you used for the analysis above?

User avatar
Master Om
Posts: 335
Joined: Wed Nov 24, 2010 9:57 am
Location: INDIA

Re: Winning after Qf6+!

Post by Master Om » Fri Oct 25, 2019 3:04 am

jp wrote:
Thu Oct 24, 2019 7:34 pm
Master Om wrote:
Wed Oct 16, 2019 5:11 pm
Okay. I am using SF Dev to analyse now. Lets see what happens.
Was SF dev the engine you used for the analysis above?
The dev doesnot find variations in appropriate time. I used mostly Allie 0.5 and Honey FD
Always Expect the Unexpected

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Winning after Qf6+!

Post by jp » Fri Oct 25, 2019 5:37 am

Master Om wrote:
Fri Oct 25, 2019 3:04 am
The dev does not find variations in appropriate time. I used mostly Allie 0.5 and Honey FD
What typical times and depths did Allie & Honey go to? (I don't know what the "FD" means.)

Others have already used SF-dev, so it's probably good that we now have engine variety.

User avatar
Master Om
Posts: 335
Joined: Wed Nov 24, 2010 9:57 am
Location: INDIA

Re: Winning after Qf6+!

Post by Master Om » Fri Oct 25, 2019 6:41 pm

jp wrote:
Fri Oct 25, 2019 5:37 am
Master Om wrote:
Fri Oct 25, 2019 3:04 am
The dev does not find variations in appropriate time. I used mostly Allie 0.5 and Honey FD
What typical times and depths did Allie & Honey go to? (I don't know what the "FD" means.)

Others have already used SF-dev, so it's probably good that we now have engine variety.
AFAIK For NNs depth does not matter much. For them the search is key. In key positions they play it intsantly where sf needs a depth more than 45 to confirm it. I feel like they play of intuition. I used allie for getting the lines. Then used Honey FD. FD means fortress detection as I felt there can b chance of a fortress and there is in the Ng5 line. Dev is fine for play. Analysis work is better with other forks.
Always Expect the Unexpected

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Winning after Qf6+!

Post by jp » Sun Oct 27, 2019 8:08 am

mmt wrote:
Sat Oct 26, 2019 10:18 pm
After a deep evaluation, 3.3 billion nodes, LC0's Qf6+ score falls a little. It doesn't see a clear win. It used 185 GB of RAM by this point.

+1.31 Qf6+ Bxf6 gxf6+ Kg8 Bd2 Qa5 Bh6 Qxc3 bxc3 b2 Bxf8 Rxf8 Rb1 Rb8 Ng5 h6 Rxh6 Nxa3 Rh7 Be8 Rxb2 Rxb2 Rg7+ Kh8 Bg4 Nc2 Bxe6 fxe6 Ra7 Rb8 Nxe6 Ne3 Kg1 g5 Nxg5 Bg6 Rxa4 Be4 Ra5 Bg6 h4 Nd1 Ra3 Ne3 Ra7 Be4 Rg7 Rb1+ Kf2 Ng4+ Kg3 Nh6 Ra7 Rg1+ Kf2 Rg2+ Kf1 Rg4 Ra8+ Ng8 f7 Rf4+ Ke2 Bh7 f8=Q Rxf8 Rxf8 Bf5 Ra8 Kg7 Ra1 Kg6 Ke3 Ne7 Kf4 Bh3 Ra6+ Be6 Rb6 Kh5 Rxe6 Ng6+ (N: 2749M, P: 6.24%, V: 0.886, Q: 0.445)

-0.85 Rh6 Rh8 h4 Qd8 h5 Nxa3 Bd2 Nb5 Rf1 Be8 Qh4 a3 Kg2 axb2 Rh1 Kf8 Nh2 gxh5 Ng4 Na3 Bf4 b1=B Rxb1 Nxb1 Rxh5 Bb4 Nf6 Nxc3 g6 fxg6 Rxh7 Rxh7 Nxh7+ Kg8 Nf6+ Kf8 Nh7+ Kg8 Nf6+ Qxf6 exf6 b2 f7+ Bxf7 Be5 Kf8 Qh8+ Bg8 Qg7+ Ke8 Qxg8+ Kd7 Qxb8 b1=Q Bg4 Qe4+ Kg3 Ne2+ Kh2 Nxd4 Qa7+ Ke8 Qxd4 Qxd4 Bxd4 Ke7 Kg3 Bd2 Be5 Bg5 Bc7 Bc1 Be5 g5 Kf3 Bf4 Bxf4 gxf4 Kxf4 Kd6 Ke3 e5 Bh5 Kc5 Bf7 d4+ Ke4 Kb4 Kxe5 c3 Bg6 Kc4 (N: 111M, P: 7.16%, V: 0.447, Q: -0.290)

-1.08 Bd2 Rh8 Rh6 Qd8 h4 Bxa3 h5 Bxb2 Kg2 Bxc3 Bxc3 a3 hxg6 fxg6 Kf2 b2 Qf6+ Qxf6 exf6+ Kg8 Ne5 a2 Bxb2 Rxb2 Nxc6 Rb1 Rhh1 Rxd1 Rxd1 a1=Q Rxa1 Nxa1 Bg4 Kf7 Ne5+ Ke8 Bxe6 c3 Ke3 c2 Kd2 Nb3+ (N: 20.9M, P: 14.92%, V: 0.442, Q: -0.369)

-1.29 Bf2 Rh8 Be3 Nxa3 bxa3 Bxa3 Nd2 Be7 Rf3 Be8 Bxc4 Qb7 Bd3 a3 h4 a2 h5 b2 Nb1 Kg8 Rdf1 a1=Q Kg2 Qaa7 Rh1 Bf8 h6 Qb3 Rf2 Be7 Bc2 Qc4 Bd1 Qd3 Bc2 Qda6 Nd2 Kf8 Rb1 Qa3 c4 Qc7 c5 Bd8 Nf3 Rg8 Nh2 Qe7 (N: 418M, P: 6.77%, V: 0.300, Q: -0.440)

-1.30 Ng1 Rh8 Bd2 Be8 Rf1 Nxa3 Bg4 Nb5 Qf2 a3 Rf3 Rg8 Rxf7+ Bxf7 Qxf7+ Kh8 Qxe7 Qc7 Qf6+ Qg7 Qxe6 Rgf8 Bf3 Nc7 Qd6 a2 Ra1 Rbd8 Qc6 Rf5 Be3 Qf7 Bg2 Rf8 Qd6 Kg7 Bh3 Kg8 (N: 1.9M, P: 2.44%, V: 0.175, Q: -0.442)

185GB RAM! How much RAM do you have?

mmt
Posts: 13
Joined: Sun Aug 25, 2019 6:33 am
Full name: Jim Fields

Re: Winning after Qf6+!

Post by mmt » Mon Oct 28, 2019 1:37 am

jp wrote:
Sun Oct 27, 2019 8:08 am
185GB RAM! How much RAM do you have?
512 GB. I'm working on an NLP project that needs this much RAM. And thanks for posting this LC0 analysis in the proper thread. I think I had some old cookies or something.

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Winning after Qf6+!

Post by jp » Mon Oct 28, 2019 6:41 am

mmt wrote:
Mon Oct 28, 2019 1:37 am
512 GB.
That's great. :) I wonder if it'll work (for others who don't have so much RAM but are patient) using virtual memory.

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Winning after Qf6+!

Post by jp » Wed Dec 04, 2019 6:55 am

zullil wrote:
Sat Oct 12, 2019 9:46 am
A much discussed position:
1r3r2/4bpkp/1qb1p1p1/3pP1P1/p1pP1Q2/PpP2N1R/1Pn1B2P/3RB2K w - - bm Qf6
The biggest failure with this position was Fat Fritz (tested by Peter in mid-October). Maybe someone with a copy can try again with long time and see whether Fat Fritz can pass this time.

Post Reply