FIBS Board backgammon forum

one vote per user .. you must login to vote .. results of the poll are shown after you have voted

Early in the match a redouble can give one side a significant lead if accepted, force an opponent to eat an earlier cube decision that has gone awry, or send the match to match point when the redouble is accepted and then redoubled. Is this a time for a redouble? GNUbg says the correct actions are

Spoiler

Cube equity analysis

4-ply cubeless equity +0.519 (Money: +0.518)
0.752 0.022 0.000 - 0.248 0.009 0.000

Cubeful equities:
1. Double, take  +0.871
2. Double, pass  +1.000 ( +0.129)
3. No double   +0.868 ( -0.003)

Proper cube action: Redouble, take


Cube MWC analysis

4-ply cubeless MWC 56.62% (Money: +0.518)
0.752 0.022 0.000 - 0.248 0.009 0.000

Cubeful equities:
1. Double, take  61.11%
2. Double, pass  62.76% ( 1.65%)
3. No double   61.08% ( -0.04%)

Proper cube action: Redouble, take

In this match my opponent accepted my redouble and to my surprise redoubled immediately after my roll of 61 which I played 8/1. I accepted and went on to win the 7-point match in one game. There are some situations where a redouble after accepting a redouble makes sense. There are some occasions when it can be an effective bluff. This was not one of them.


Cube equity analysis

4-ply cubeless equity -0.513 (Money: -0.515)
0.251 0.005 0.000 - 0.749 0.022 0.000

Cubeful equities:
1. No double   -0.411
2. Double, pass  +1.000 ( +1.411)
3. Double, take  -0.939 ( -0.527)

Proper cube action: No redouble, take (27.2%)


Cube MWC analysis

4-ply cubeless MWC 36.36% (Money: -0.515)
0.251 0.005 0.000 - 0.749 0.022 0.000

Cubeful equities:
1. No double   39.07%
2. Double, pass  76.57% ( 37.50%)
3. Double, take  25.06% (-14.02%)

Proper cube action: No redouble, take (27.2%)

[close]

Enter the ID's below into GNUbg or Snowie to further evaluate your answer:

Position ID: bduBAAx43j0AAA Match ID: UQngAAAAAAAA

If the board image does not appear in the question above, you may view it in the attached PNG file below.


Life was simple before World War II. After that, we had systems.
-- Rear Admiral Grace Murray Hopper (1906 â€" 1992) was an American computer scientist and pioneer in the field, she was one of the first programmers of the Harvard Mark I calculator, and she developed the first compiler for a computer programming language, CoBOL.

both Zorba in shouts and dorbel in an email have suggested this problem needs a rollout for further evaluation .. i've been real busy lately and don't have the time at the moment to do one, but will .. i ran all the evaluations at 4-ply for situations and thought that would be enough, but i am very curious about evaluating positions like this with rollouts versus a 4-ply evaluation .. maybe someone can explain in words i can finally understand what exactly one finds with a rollout and why this position seems to suggest it needs one, because, frankly, i am beginning to think i am clueless about rollouts versus evaluations

meanwhile, i look forward to someone adding come clarity regarding rollouts with GNUbg or Snowie, i.e. what are they? when to use them? what signals when a rollout may be useful? etc. 

A rollout is used particularly when there are dramatically different MWCs/equities across the first roll (occasionally a rollout will also be done for the second player's response roll). It shows in more detail exactly where the MWCs/equities etc in a 4 ply evaluation are coming from eg - if I hit then.... , etc.
It's useful not so much that it will give a different result to a 4 ply analysis (it won't of course - summing across the rollout will give the same outcomes) but that it will explain where those values are coming from. In general a rollout is particularly useful when you have a simple bistate to analyse eg I hit / I miss. It follows then that in simple positions (the simplest of all being a post contact race) a rollout is of very little use.

Yikes!  So many words!   

socksey



Dear Lord,  So far today, i am doing all right.  I have not gossiped, lost my temper, been greedy, grumpy, nasty, selfish, or self-indulgent.  I have not whined, complained, cursed, or eaten any chocolate.  I have charged nothing on my credit card.  But I will be getting out of bed in a munute, and I think that I will really need your help then. 

It's good to see Spielberg attempting parody, a rarity in backgammon fora. However, for me, it is too obviously balderdash to be truly funny. I would have liked to see one or two facts to confuse the unwary into thinking that it was serious. Still, nice try!

well, count me amongst the unwary .. if balderdash, spielburg's overview was beautifully written balderdash .. i am still clueless enough about really understanding rollouts that it almost made sense to me

i was hoping you, dorbel, and/or Zorba (the birthday boy) could offer up an explanation and answers to my questions .. i still haven't had time to run the rollout for this position to see what it is that makes it useful .. so, i'm really still clueless about what you guys saw in this position and about using rollouts in general .. i have used GNUbg for well over a year now, probably more seriously than the average user .. i know for certain i still have something to learn about rollouts and have yet to see or understand a well written explanation of how to use them

"So many words!"? now, socksey, you gossipy, hot tempered, greedy, grumpy, nasty, selfish, self-indulgent, whining, complaining, accursed, chocolate eater you, have a little patience, sweetie  .. you of all people know it takes some of us 100 words or more to say hello, e.g. moi

An evaluation is a bot's attempt to estimate the likely outcome of a play or a cube action. A rollout on the other hand, means that the bot plays the position to the end of the game a preset number of times. From this it derives statistics that are recognised to be more accurate than the evaluation, assuming that the operator has used settings that are appropriate to the position.
You need to roll out this position because Snowie and Gnu evaluations routinely over-estimate the winning chances of the side holding a 22pt anchor, but evaluation alone should never, in my opinion, be presented in an article as if it was fact. It isn't difficult to learn how to do rollouts, nor is it time-consuming, as the program will quite happily crunch the numbers while you sleep.
Far from being a simple take, this position is in fact a very thin take indeed. Moving one checker from the 8pt to the 7pt is, in my opinion, enough to make it a pass. Thus we can see that this is a very useful reference position, that can be used to evaluate similar positions over the board, as it is a marginal take/pass. A little better for the defender, take, a little worse, pass. It is worth memorising and you can do this quite easily by playing it a few dozen times, until you can set up the position from memory.

http://www.bkgm.com/gnu/AllAboutGNU.html
http://www.gammonlife.com/gnu/index.htm

The first link takes you to Albert Silver's excellent guide to using gnu, the second to the page of our very own Zorba's guide. I have used the first one a lot and by reputation, Zorba's is also excellent. You should find all that you need there.