Based on that experience, I started a few days ago playing some money sessions by myself against the computer. I play about 10 games (which goes really quick) then peruse the errors after.
Stick has recommended keeping a working file of blunders based on these games, and Robertie recommends categorizing your errors and printing them out  maintaining a working binder of errors.
All good advice, but I am not quite ready to go that extreme yet.
Here are some 3rd roll errors I have in recent sessions against XG.
***************
32 to play. 63S52S32??
is eXtremeGammon score: 1 pip: 160  
Unlimited Game Jacoby Beaver  
pip: 158 score: 6 is Bill $$  
XGID=aaECAdDdAeA:0:0:1:32:6:1:3:0:10  
to play 32 
1.  XG Roller++  18/16 13/10  eq: +0.093  
 
2.  XG Roller++  18/13  eq: 0.022 (0.115)  
 
3.  4ply  6/3* 3/1*  eq: +0.056 (0.037)  

eXtreme Gammon Version: 2.02
I moved 18/16 running a man to the midpoint.Better was 18/16, 13/10.
My play leaves 7's and 9's to hit. That's 11 shots, though XG may forego the hit on 33. Let's call it 10 shots.
The other play leaves 13 shots, but 22 44 and 31 are good shots anyway so let's call it 8 shots. So with some sort of duplication at work, the direct shot costs less than the indirect shots.
And, of course, the bot's play makes a point. Making a point is always good in the early game, no matter where it is.
****************
52 to play. 43Z52H52??
is eXtremeGammon score: 3 pip: 160  
Unlimited Game Jacoby Beaver  
pip: 164 score: 10 is Bill $$  
XGID=bECAdDddaAA:0:0:1:52:10:3:3:0:10  
to play 52 
1.  XG Roller++  Bar/23 13/8  eq: 0.117  
 
2.  XG Roller++  Bar/20 13/11  eq: 0.118 (0.001)  
 
3.  XG Roller++  Bar/20 24/22  eq: 0.139 (0.022)  
 
4.  XG Roller++  Bar/20 6/4  eq: 0.150 (0.033)  
 
5.  Book^{1}  Bar/18  eq: 0.208 (0.091)  
 
^{1} Generated by GameSite 2000, Ltd on 2/26/2011 using eXtreme Gammon 2.00 Analyzed in XG Roller++ 
eXtreme Gammon Version: 2.02
I played Bar/18. This is a terrible blunder. I place my checker on a point he wants to make and diversiy his numbers. 2's and 4's cover while 1's and 6's hit. Some numbers hit and cover. Too much, and why?Two plays are better.
One is the "normal" looking Bar/22, 13/8. This doesn't do much except stay out of harm's way but remaining on the deep and ace duece points. 13/8 isn't great since the midpoint is already unstacked but once you enter on the duece what other five is there? Well, there is my ridiculous Bar/21/18 but let's forget I did that!!
The other is the tactically oriented Bar/20, 13/11. This places your checkers where you want 'em. It is a bit blotty but the blots in my outfield are sort of a mirage. These guys are immune for one roll since any indirect hit by XG gives me fours to hit back from the bar and fives to anchor and perhaps combo two's as well to hit back in my outfield. So barring a joker dance these guys are really safe this turn. Staying on the five point is a bit risky but the blot on his four point mutes the danger somewhat. There are only a handful of hit and cover numbers. Barring that I should get an opportunity to make a good anchor next turn or at least make an offensive point.
*************
21 to play from the air. 64R52H21??
is eXtremeGammon score: 3 pip: 160  
Unlimited Game Jacoby Beaver  
pip: 168 score: 11 is Bill $$  
XGID=bECcEadeAA:0:0:1:21:11:3:3:0:10  
to play 21 
1.  XG Roller++  Bar/24 13/11  eq: 0.176  
 
2.  XG Roller++  Bar/22  eq: 0.233 (0.057)  
 
3.  4ply  Bar/23 6/5  eq: 0.185 (0.008)  

eXtreme Gammon Version: 2.02
I played Bar/22, inching up to influence the outfield and splitting to try and make an anchor.Better was Bar/24, 13/11.
I am not in urgent need of an anchor here. He has made no points yet and in fact has a stacky position. He would love to unload his stacks upon my pretty little head. Better to fight a stacky attacky position with an anchor than to 'spread the legs.'
Besides those defensive points, Bar/24, 13/11 also unstacks my midpoint to an effective spot to develop my forward position. The 11 point is only vulnerable to 64 (which as a bonus 64 is duplicated on his side of the board). meanwhile the 11 point builder aims at both the bar and the 5 point.
This play seems pretty clear to me now. Hopefully I will get it right in the future.
*****************
53 to play. 62S22A53??
is eXtremeGammon score: 2 pip: 159  
Unlimited Game Jacoby Beaver  
pip: 159 score: 3 is Bill $$  
XGID=bECAcDbcAcbA:0:0:1:53:3:2:3:0:10  
to play 53 
1.  XG Roller++  18/10  eq: 0.251  
 
2.  XG Roller++  18/13 11/8  eq: 0.282 (0.031)  
 
3.  XG Roller++  8/3 6/3  eq: 0.316 (0.065)  

eXtreme Gammon Version: 2.02
I made a point 8/3, 6/3. While it would be nice to make the bar point this just seems too unlikely to work. I am staring down a triple shot and standing on a point he would actually like to make. At first glance this looks like a distraction play but XG already has the 4 point made so building either the bar or the 5 point starts to fill in a budding prime. It looks in most variations I get and if I don't perform from the bar I am in big trouble already.
Better was 18/13, 11/8. I suppose this way I hope to escape one man. If I am not hit, then I am good shape to make a point somewhere next turn with a playable game. There is also some duplication at work  63 64 are both good shots on his side of the board  but I don't think this is the main reason.
Also not the wimp inspired 18/13, 11/8 is a reasonable if not entirely correct play. This escapes the back man for sure by leaving no blots but at the cost of a stacked position. XG is building structre quickly so I will need to roll well from here for this to work but note it is still superior to my play, which 'leads with the chin.'
Good problems, Bill. For 64R52H21, I would have played bar/23 6/5, which I notice you did not bother running through XGR++. I'm curious as to how well it compares. Perhaps a full rollout is needed.
ReplyDelete