White - Pips 158

Black - Pips 167
Black to Play 6-3

What’s your play, and why?

This is a rhetorical question. I could play 13/7 8/5, though if being ahead in a tournament i would play the prudent 24/18 8/5. Checking however the analysis, to my big surprise 13/7 8/5 is on top of the list.

Out of curiosity i tried the followup into the middle game 5 times, and those positions showed up:

[img]http://s6.************/ms8jpeohd/Screenshot_from_2013_07_17_09_08_16.png[/img]
green shot screen capture

[img]http://s6.************/yjwf07j3l/Screenshot_from_2013_07_17_09_28_22.png[/img]
small screenshot software

[img]http://s6.************/4t9aefy41/Screenshot_from_2013_07_17_09_33_44.png[/img]
open source screen capture

[img]http://s6.************/lv24gjcz5/Screenshot_from_2013_07_17_09_38_12.png[/img]
image hosting software

[img]http://s6.************/46adox181/Screenshot_from_2013_07_17_09_43_03.png[/img]
image ru

Adding the equities of the 5 positions, i arrive at .256 + .437 + -.266 + -1.059 + -.080 = -.712, divided by 5 give a disappointing average loss of .14. I guess i was very unlucky in the 4th game. In anyway, it will create situations in which also an advanced player is going to make big mistakes.

I thought you’d like this one. I read your post saying you’ve been slotting your early 6s of late (albeit, mostly for entertainment).

Any idea why 13/7 8/5* is correct here?

What about running, 24/15, or hitting and splitting, 24/18 8/5*?

Mike

Running is good for a racing game, decreasing gammon loss equity.

Hitting and splitting is the balanced play, that i have learned from the Robertie books. And this is the right play for beginners and intermediates, but i try to make myself believe that this is not necessarily so for the more advanced player.

If nature was all about balance, the world would only be inhabited by one-celled organisms, and evolution would not be possible. But biological systems get sometimes into crisis, and out of the resulting chaos the system might be elevated to a higher complex stage. Translated to backgammon, i like to create an imbalance in the very beginning stage, when not so much harm can be done. If racing considerations would be so important in the initial stage, than there would be no routine hit on your 5 point, as the opponent is favourite to hit back. Even if a starting 52 is played 13/8 6/4 the equity loss is just 0.05. Another argument is that running and balanced play support strategies in which the aspect of backgammon as being primarily a racing game is emphasized. This sounds like rebelling against our revered moderator, whose insight in the backgammon universe is zillion times superior to mine, but we are all very small compared to the universe, so i desire to keep open the possibility that there is dark energy that i just happen to detect, though the odds are against me.

Now why would in mentioned case 13/7 8/5* stand out? One thing is that the 8 point builder loss is made good in some sense. Another thing is that a first move of 64, 63, 62 results in an equity loss of -.09 in Gnu supremo, which is an error magnitude that many of us would like to see within the margin of many of our cube decisions.

Concerning the bot evaluation of starting moves i would like to pose the question, to what extent even a bot like XG is capable to do a reliable calculation, as the game can flow into numerous directions. Secondly, suppose that XG is doing its job very well also for the starting moves, doesn't the fact that there are different playing styles, levels and quirks, play down the value of this evaluation greatly?

I'm not questioning the 8/5* part, it's more the 13/7 part. Slotting 2 points you want to make at the same time is usually not that good of an idea, unless there's nothing better to play. Here, 13/7 gives another direct return shot. Is it worth it if you're not hit? Most of the time, you won't cover both slots. If you end up covering the 5-pt slot with the blot on the barpoint, then slotting the barpoint might have been unnecessary risky.

Also, with the 24/18 8/5*, it's a balanced play, but it's not "quiet balance", it's "provocative balance", since you hit loose in your home board. It's the first act of war of the game, so chaos and evolution will often arise from such a "balanced play".

Edit: Some of my humble reasoning might be wrong, I don't pretend being right 100%, it's just to continue with this interesting discussion.

According to my line of thought the emphasis here is not so much on slotting, as on creating havoc. Covering is a welcome byproduct.

But i think it is time to put the 'rampage' play and the 'provocative balance' play to the test, so that any linguistic prejudice is ruled out.

Giant1 played 13/9 13/7 as opening move against Giant2. Playing 12 moves for 15 games brought 4.24 equity points of winnings and 4.36 points of losses.

Next i have put Giant1 (me ofcourse) against an advanced player doing the same opening move for 10 games, resulting in .966+.034+.579+1+1+1.964+1+.804= 7.347 on winnings and .359+1= 1.359 losses, totalling 5.988 points, giving a positive equity of .6 per game. Like in my post above, i show below the resulting positions into the middle game.

I have the following request to Mike, and maybe he can split these night jobs with whomever offers himself.
Let XG do rollouts on the opening move 13/9 13/7 with the parameters (4-ply, 4ply (for expert), huge, 1296) with the following player settings: expert-expert, expert-advanced, expert-intermediate, advanced-intermediate, advanced-beginner. In case you set yourself to this task, then post the outcomes in a new thread.




windows screenshot

[img]http://s6.************/627s17w4x/Screenshot_from_2013_07_18_01_35_50.png[/img]
upload a picture

[img]http://s6.************/xe312k0vl/Screenshot_from_2013_07_18_01_38_47.png[/img]
screenshot tools

[img]http://s6.************/9pnj7a4c1/Screenshot_from_2013_07_18_01_41_58.png[/img]
image upload no limit

[img]http://s6.************/l3a2iheup/Screenshot_from_2013_07_18_01_45_41.png[/img]
image upload no ads

[img]http://s6.************/92omhr7g1/Screenshot_from_2013_07_18_01_51_36.png[/img]
image hosting imgur

[img]http://s6.************/ibqsrvgc1/Screenshot_from_2013_07_18_01_53_00.png[/img]
picture sharing

[img]http://s6.************/g8gdk7gj5/Screenshot_from_2013_07_18_01_55_33.png[/img]
image hosting 15mb

[img]http://s6.************/4kmbpnre9/Screenshot_from_2013_07_18_01_58_42.png[/img]
image hosting

[img]http://s6.************/v73dsdkz5/Screenshot_from_2013_07_18_02_02_21.png[/img]
image hosting 20mb

[img]http://s6.************/uiuj9fm9d/Screenshot_from_2013_07_18_02_07_45.png[/img]
how to take a screen shot

[img]http://s6.************/cu2shtai9/Screenshot_from_2013_07_18_02_12_17.png[/img]
windows screenshot

This is a very tough question. There is no doubt that the bots are right when they split, 24/18, with so many early 6s. That’s the big lesson.

After you are convinced of that, it is fun to look at a few of the exceptions. When they come on the second roll, that gives an extra impetus to examine them, because you are sure to see them over the board.

This is a valid point, but does it give you the tools to recognize over the board that this is one of the very few times where splitting is wrong? Probably not.

Perhaps it is best not to worry about the exceptions. The error made by splitting off the 24pt point will usually be small. In the case of 54-split 63, playing 24/18 8/5* is only an error of 0.025.

Bruce Becker wrote Backgammon for Blood, one of the worst backgammon books of the 1970s, indeed, of all time. Its central tenet was to slot every opening where you could not make a point, including the 6s.

Trust me, that’s bad advice.

Mike

Although I would not mind seeing the results of such an endeavor, this one is not a job for me. As it is, I am burning too many cycles on positions where I don’t know the correct play!

As well, XG, unlike GnuBg, does not have the capacity to do rollouts where opposing sides use different strengths.

Love to see your GnuBg results.

Mike

Even at gammon-go, playing down from the midpoint, 13/9 13/7, with an opening 64 is a loser. See this 10k rollout by Ian Shaw. If ever there were a time for the slot, gammon-go would be it.

You can, however, make your case in other positions. One outstanding example comes in reply to an opening 64. After your opponent splits, 24/18 13/9, you can hit when you roll 31! The proof comes by way of a 46k rollout by Neil Kazaross.

There are plenty of other examples where a close choice can be resolved in favor of a play that increases “complexity.” For the advanced player, they give an opportunity to create havoc without sacrificing equity.

Mike

First choice 8/2 6/2? This is no counter argument, but an escapism. I mentioned that i tried supremo for 15 games resulting in a draw, and .03 to me is no difference, and much better than gnu's evaluation for the opening move. Where is XG, and why do you not bother about the difference in playing level?

At gammon-go? Absolutely. Remember, at gammon-go, it’s the Crawford game, and you need 2 points.

Dmitriy Obukhov and Paul Weaver have authored an article on opening 64 that will be published in the upcoming issue of Primetime Backgammon. They give lucid explanations why 64 might be played Point = 8/2 6/2, Split = 24/18 13/9, or Run = 24/14 depending on match score. In unlimited games (money), you pick. All three are essentially tied in this 46k rollout from the XG Opening Book. One play that is not correct, however, is Slot = 13/9 13/7.

It is hard to assess the value of your rollouts. With only 15 trials, they are anecdotal, and should not be taken as representative of a general result. Nevertheless, hand rollouts such as yours are extremely valuable for learning about a position. So I say, keep it up!

In Backgammon Openings, the excellent book by Nack Ballard and Paul Weaver, there is an Error Scale whereby the authors describe any plays within 0.01 of best as “tied.” Within 0.02 is “very close;” and 0.03, “close.” More than 0.03 is a “mistake.”

For checker plays, I would like to be within 0.015 of best. I think you are giving away too much to allow 0.03 and above.

I don’t feel that I am good enough to judge when to deliberately misplay a move because I believe a particular opponent will respond with a particular misplay of his own.

Mike

-

Sorry Mike, I missed a couple of your posts. Today i have not much time at hand, so tomorrow i will respond.

I am not expert at neural net programming. Someone who is might be able to give a better answer than me.

I know that the poor play bots exhibit when they try to roll a prime around the board means that they generally value backgames less than they should. Does this mean that a bot will misevaluate a risky slotting play such as the opening 64 you are toying with?

I think yes, but the effect is small. With a large number of trials in a rollout, backgames will be represented. The bot will play some, but not all, or even a majority of them, badly. This will skew the resulting equities in the rollout. In the larger scheme, however, the number of backgames will be very small. That means that the skew in equities will also be small.

When bots learn to play backgames better, there will be a change in the ranking of some early game moves that are close. Think about opening 64, for instance. The ranking of the top three plays (split, point, run) might easily be permuted. For other plays, such as 64-slot, however, which is a whopper that trails the top three by 0.08 or 0.09, I do not expect to see any radical revision.

This must certainly be true.

I think, however, your phrasing tends to exaggerate the effect. I prefer to say “somewhat” rather than “greatly.” Cube tendencies, more than styles of checker play, might be an area that could be profitably explored.

One reason we have not seen an attempt to build something like this into our bots is because the programming is so different than neural-net programming. The current bots do not understand, and cannot identify, “playing styles.”

Furthermore, any adjustments that were made might do more harm than good.

Mike

Assuming an unlimited game with Jacoby I would defiantly be hitting with 8/5, I couldn't allow my opponent to grab that next time and it is the most important point on the board at the moment.

That leaves the 6 and the choices are 24/18 or 13/7.
24/18 is looking to make an advanced anchor or race but that is not where the battle is at the moment.

I prefer 13/7 which is much more aggressive. Here I am looking for an early blitz which could lead to an early double. The worst that can happen is I get hit twice but as Whites board is so weak, this is no problem to me.

I'm just a beginner in match play so I would be interested to know the answers for DMP, GG, GS.

For DMP where the first person who wins the game wins the match, I might be better to race with 24/15.
For GG where I want to win a gammon, I would play 13/6 8/5*, same as cash.
For GS where I don't want to lose a game I might try 8/2 6/2 which is very safe but that just doesn't feel quite right to me.

White - Pips 158

Black - Pips 167
Black to Play 6-4

Staying with unlimited games for now (Jacoby enforced and beavers allowed), does the same thinking apply when your opponent opens 54-split, and your roll is 64?

Mike

I just noticed the cube is not centered in the diagram above.

Of course, it should be.

Mike

Now I can't hit with 64. Aggression here is like the boxer who lets down his guard to take a big swing - he's likly to get one full in the face.

I would play 24/14 which I think is second safest but it moves a piece in to a position to do something good next turn.

Here are my thoughts on each of the moves:

My choices are:
24/14 - Nothing too bad with running in a race and is a nice builder for a future prime. I won't be happy to be hit but it is no disaster.
24/18 24/20 - This is just inviting white to launch a blitz.
24/18 13/9 - Lots of blots for white to hit. Don't like this.
24/20 13/7 - Slotting two valuable points but black is able to hit with almost any roll (55 and 54 miss).
13/7 13/9 - Aggression becomes foolhardiness.
13/3 - Puts a blot on a point that I don't need and I'm very unhappy if hit.
8/2 6/2 - Very safe and makes a point, but not a very valuable point at the moment.

I check in the dark. Probably written during the Vietnam era. Chis Bray is not entirely unappreciative about the book. Anyhow, i restrict hippielike behaviour only to the first move(s). Having read Bill Robertie's problem commentaries, it has happened that a once rejected approach has been revoked. I could imagine, that an old strategy has been declared inferior at the arrival of the bot, but regains its validity in an era of perfected bot development and it's resulting extremely improved checker know-how.

Though i respect your intellectual capabilities, I am not so impressed with statements like ¨trust me¨, ¨i think¨, ¨i prefer¨, ¨might¨.
A onetime adjustment that provokes a lesser opponent into a course of events which will cause him to make big mistakes, will do more good than harm. Unfortunately, i am not able to come up with some bot-evidence, as it is likely that gnu also offers no facility to do rollouts between different player levels. At the moment we have to rely on common sense, experimentation and handrollouts, and i hope i will get some positive feedback in the future

In this post, Chris call the Becker text “awful,” and decries the decision of his American publisher to use the title for his own work.

Check again. For starters, reread my post from this thread. It says plainly that GnuBg does have that capability. Then review this GnuBg rollout of opening 64 where said capability has been employed. Note that Player 0 is set at “expert,” while Player 1 uses “world class.”

If a similar rollout were to show 64-slot to be inferior, would you accept the result? If not, is there any evidence you would accept? Is your opinion subject to change?

On this, we can agree.

If you do continue accumulating data from hand rollouts, may I suggest that you play an equal number of games with a conventional 64-split opening? That way, you could compare the two to see whether 64-slot was yielding more equity.

Better still would be to find a partner, and play it as a prop. Place a small stipend on the games, and alternate openings. On his turn, let your partner open with 64-split, while you play 64-slot on yours. Keep playing—and paying—until one of you is willing to concede the argument.

Mike

Sorry, i read a little review at a bookselling site.

Please, can it tomorrow, or the day after.

Quote:

If a similar rollout were to show 64-slot to be inferior, would you accept the result? If not, is there any evidence you would accept? Is your opinion subject to change?

To be square, i doubt it. I find it big fun. I suspect that in case the slot is inferior, the equity difference will not be much. My argument will be that the bot will not take into account quirks and mistakes because of stress and fatigue. On top of that, in case you lose from an inferior player, you have some unspoken excuse, and in case you win your superiority is beyond all doubt.

The only partner i play occasionally with is a beginning player. It would be a longterm project. Better to make it a collective project. This thread runs well, with hundreds of attentive and enthousiastic viewers.

You are right.

This point has been discussed at length at BgOnline.org by world-class players such as David Rockwell, Nack Ballard, Neil Kazaross, and Stick. It’s a gambit where you sacrifice equity x in the hope to gain more than x by subsequent opponent misplays.

The only disagreement we have is whether the concept should be applied to an early slot of the bar point. The experts cited above all agree that x should be on the order of 0.01, or at most 0.015, for the gambit to be worth trying. I have been trying to convince you that slotting the bar generally gives up too much. In the specific case of opening 64, slotting is a whopper that gives up about 0.09. A hypothetical, lesser, opponent is going to have to be a pretty bad player for you to be willing to hand him that much equity at the start of every game.

Knowing of your interest in slotting an early 6, I began this thread with 54-split, 63. It’s a position where you can have your cake, and eat it, too. Since it looks like neither one of us will the convince the other on the broader issue of opening 64, let’s just agree that we’ll both be slotting—and winning—on that one.

Mike

You have mentioned that you play tournament in San Francisco. Let everyone open with 13/9 13/7 for the one tournament, and 24/18 13/9 for the next tournament, just ignoring player's level, and note the results (and be objective). Though it is very unrealistic, i would like to propose this in general, and show up here with the results.

Question for Mike. What settings do you like me to use for the rollouts?
[img]http://s6.************/m5w70r629/Screenshot_from_2013_07_21_14_37_34.png[/img]
upload

.