Deep Snake Cube - Revisited

You may remember the position that Bob Koca originally posted ( https://www.bgonline.org/forums/webbbs_config.pl?noframes;read=219822> )

I�m always happy to see a position like this from a real match, because BGBlitz has a clear advantage in these types of positions.

But how do you demonstrate something like this? A rollout doesn�t provide a definitive answer if Bot finds Move_1 to be better and Bot B finds Move_2 to be better. I know of only two possibilities:

You can identify it through logic. The problem, however, is that one person might see it one way, while another might see it completely differently. A consensus among several grandmasters is helpful here (though they aren�t always available).
If one side has practically no decisions to make, a rollout reveals the differences.

I had therefore created a rollout, but Bob Koca rightly pointed out that errors can be made with the cube. To rule that out, we�ll turn it into a DMP position so that one side is trivial to play.

Position_1

XGID=----a--BBBBB---AA--A-cdgB-:0:0:1:00:6:6:0:7:10

Stick�s rollouts (https://www.bgonline.org/forums/webbbs_config.pl?read=219867>) couldn�t be used for this reason, so I rolled it out with XG using the commonly used settings: 3-ply/XG+/normal:

White is Player 2

score: 0
pip: 59

1 point match

pip: 188
score: 0

Blue is Player 1

XGID=----a--BBBBB---AA--A-cdgB-:0:0:1:00:0:0:0:1:10

Blue on roll, cube action?

Analyzed in Rollout	No double	Double/Take
Player Winning Chances:	64,38% (G:4,48% B:0,12%)	64,38% (G:4,48% B:0,12%)
Opponent Winning Chances:	35,62% (G:26,82% B:8,48%)	35,62% (G:26,82% B:8,48%)
Cubeless Equities	+0,288	+0,288
Cubeful Equities
No double:	+0,288 (0,000)	�0,025 (+0,263..+0,313)
Double/Take:	+0,288	�0,025 (+0,263..+0,313)
Double/Pass:	+1,000 (+0,712)

Best Cube action: Double / Take
Rollout details
1296 Games rolled with Variance Reduction. Dice Seed: 59890281 Moves: 3-ply, cube decisions: XG Roller

Double Decision confidence:		50,0%
Take Decision confidence:		100,0%
Duration: 1 minute 24 seconds

eXtreme Gammon Version: 2.10, MET: Kazaross XG2

Larger filters typically help with these types of positions, and I also set XG to its strongest possible level: XG++/XG++/Gigantic. That�s why the rollout took 23 hours instead of 84 seconds and yielded:

White is Player 2

score: 0
pip: 59

1 point match

pip: 188
score: 0

Blue is Player 1

XGID=----a--BBBBB---AA--A-cdgB-:0:0:1:00:0:0:0:1:10

Blue on roll, cube action?

Analyzed in Rollout	No double	Double/Take
Player Winning Chances:	74,30% (G:4,64% B:0,13%)	74,30% (G:4,63% B:0,13%)
Opponent Winning Chances:	25,70% (G:17,86% B:5,31%)	25,70% (G:17,85% B:5,29%)
Cubeless Equities	+0,486	+0,486
Cubeful Equities
No double:	+0,486	�0,013 (+0,473..+0,499)
Double/Take:	+0,486 (0,000)	�0,013 (+0,473..+0,499)
Double/Pass:	+1,000 (+0,514)

Best Cube action: No double / Take
Percentage of wrong pass needed to make the double decision right: 0,0%
Rollout details
1296 Games rolled with Variance Reduction. Dice Seed: 59890281 Moves and cube decisions: XG Roller++ Search interval: Gigantic

Double Decision confidence:		50,2%
Take Decision confidence:		100,0%
Duration: 23 hours 01 minute

eXtreme Gammon Version: 2.10, MET: Kazaross XG2

That�s a whopping 10% increase in winning chances, meaning the commonly used setting is way off the mark here. Let�s see what BGBlitz has to say about this....

By mistake, I had started BGBlitz with a 1-ply rollout, and BGBlitz already reaches nearly 74% with that, taking 34 seconds instead of 23 hours..... . BGB_ro_1ply

The 3-ply rollout comes in at

BGB_ro_3ply

Since I thought 4-ply would certainly be useful for this difficult position, I specifically added 4-ply to BGBlitz. That results in:

BGB_ro_4ply

so there�s no significant change compared to 3-ply.

Summary:

BGBlitz 1-ply plays this position almost as well as XG with XG++/Gigantic
BGBlitz 4-ply wins 10% more games than XG at the strongest possible setting
At the same level (3-ply), BGBlitz wins 20% more games.

If you�re skeptical, just play from this position using XG and BGBlitz in their free versions, and the difference will become clear.

You can find the data here: https://www.bgblitz.com/download/Doc/Deep_snake_cube.zip

Post Response

Your Name:
Your E-Mail Address:
Subject:
Message:
If necessary, enter your password below:
Password: