If you look at scores where both players are more than 3-away and round the G11 figure to the nearest whole percent, you will find that G11 and HW are virtually identical. Of the 28 non-tied scores where both players need 4 or more, both MET's give the exact same winning chance in 25 of them. The only exceptions are: 4a 6a (G11 gives 65%, WH 64%), 5a 11a (G11 80%, WH 81%), and 8a 10a (G11 61%, WH 60%).

Some differences exist at Crawford and 2-away scores along with the well known 3a 4a score. But at scores where both are farther away, the two tables are almost interchangeable.

I think someone once rolled out a huge number of matches where GNU (using HW) played GNU (using Snowie's MET) and the side using Snowie's won by a statistically significant margin.

I decided that Snowie's MET was worth a try and, with a little work, was able to come up with a Neil's number system that is either right or within 1% at all scores in a 15 point match (excepting Crawford, Post-Crawford, and 2a scores). So it's pretty accurate and yet simple enough to use OTB without burning your brain out.

So my advice is to pick an MET and figure out a Neil's number system for it. Remember too the following good advice that GOLer Ilia Guzei once mentioned to me: If your opponent plays too conservatively then his gammon chances might well be in the 20% range which means you should use the HWMET when figuring your opponent's winning chances at a given score.

But if you're up against a shark like Malcolm Davis, Kazaross, Senkiewicz, etc. use the Snowie figure.

-- John O'Hagan

I use g11 because it was rolled out and should be accurate enough. I truly hope that the GNU programmers make g11 or mec26 the standard rather than Zadah's.

I fully believe that if bots get about 27% gammons, today's better players also do.

-- Neil Kazaross

I also use Zorbas g11 as my default MET. The reason I don't use my own MET is that it's only valid for scores upto 5a5a, the rest of the table having been generated by Mec26. I'm not sure whether this is truly a problem or not, as most met's look pretty similar as the match scores are further from finishing.

Another concern I had about my MET was that we humans don't play as well as Gnubg 2-ply, therefore does it make sense for us to use a met that is based on a playing standard that the great majority of players never get close to? I don't think so.

About a year ago I started work on developing a theoretical MET based on the type of calculations and methods described by Tom Keith (I'm not good at doing links, but you'll find it on his site if your interested). My MET is designed to account for differing win rates and gammon rates at different scores and also considers cube liveliness. With the right inputs it actually does a pretty good job of mimicking g11. My work on this project has been in hiatus for a while and I haven't polished off some rough edges yet. What I hope to achieve is a MET that is theoretically consistent with empirical rollouts and is designed to reflect a playing strength somewhere between Gnubg's 0-ply and 2-ply. Naturally, the 2-ply rollouts I've already done are useful benchmarks in this regard.

-- Ian Dunstan

I average the 2 tables. Ian's 2 ply rollouts with others 0 ply. I then use the simple to remember 32% for -2-1C and continue on from there where extrapolation is needed once we are outside your 5 pt match rollouts.

-- Neil Kazaross

"I average the 2 tables. Ian's 2 ply rollouts with others 0 ply."

This is kind of what I had in mind for my theoretical met program. Firstly, find a set of inputs that do a good job of generating a g11 look-a-like (and/or your Kaz/Shaw results); secondly, do the same for inputs that will mimic my 2-ply table; thirdly, average the inputs and see what results. I was also aimimng for a 1a2aC entry of 32%, and I could easily tweak my inputs to get this figure later.

-- Ian Dunstan


   "Is the Ortega/Kleinman MET really way off?" started 111505 by Klaus Evers  

As an Intermediate player, I never made up much thoughts about METs. My first introduction to that subject came with the fine book of Ortega/Kleinman: "Cubes and Gammons near the end of a match". That was an eye-opener for me, and from that point on I kept the current match score into account while playing matches; espec. at -2; -4 of cause. I went on using the Ortega/Kleinman MET while analysing my matches with GnuBG. I did not care for things like MEC26, g11 and so on, since the O/K MET was in line with the empirical driven chart of Kit Woolsey; which sounds resonable to me. But, lately I had a little discussion with an expert on FIBS (constantly way over 1950+), and he looked at me like I came frome the stone ages telling him that I _still_ used that old chart. (Sth. like using postcards in times of email, mobile and SMS). Another player (good, but not WC) said it didn't matter much, as long as you at least use one at all and stick to it. So, how far off am I while using this boring, out-dated chart? what are the main differences, and do they matter at scores like -2;-3 / -2;-4; / -2;-5, which I am mostly interested in? Can the difference between the charts be described in EMG equity or rating points? Just yesterday I turned on the MEC26 in my GnuBg version, is that better/ best? Can I do some rollouts within GnuBG, giving White the Ortega MEC and Black the MEC26 or g11 ? Thanks for helping me out of this darkness and confusion. -- Klaus Evers Choice of MET is "important" in the sense that if you can get anything at all out of a MET, you want it to be as accurate as possible. Every driblet of equity counts, and saved driblets can mean many monetary units over your lifetime. It's "unimportant" in the sense that the good player said. Both Ortega-Kleinman and a more modern table (e.g. g11) will steer your plays and cube actions in the same direction. You need to know when (at what scores and in what types of positions) to be aggressive, and when to be conservative. For example, how is a score of -3 to -2 different from pure GG (-2 to -1 Crawford)? What are examples of how you play differently at those two scores? I submit that the answers won't depend much on the specific MET. Though the specific MET will occasionally point you to different decisions for the same score, that won't happen too frequently or with too great an equity difference. Much more important is to be able to look at the score and "feel" the strength of the bias towards conservativism, aggression, simplicity, or complexity. First you want to get to the Zen-like state of having the right feelings: to instinctively or involuntarily start steering your play in the right direction. The beginning of such a state is to start slotting and building at GG, versus splitting and running at GS. Then you can make refinements: adding more reference positions to your memory; trying to figure out how much to change strategy against much stronger players (e.g. you'll play simpler positions better than complex positions, but complex positions are more volatile, so, how do you play opening 51 at -3 to -3 against Neil Kazaross? how does your take-point change in a straight race? in a holding game? etc.); refining strategy against weaker players; and there's really no end to refinements you can make. So if you've already memorized a Woolsey-Heinrich-like MET, you can keep using it for a while, with no really bad consequences. But if you want to be as accurate as possible, ditch Ortega-Kleinman/Woolsey-Heinrich for g11 or for one of the built-in bot MET's. Once you start looking deeply enough into positions for Gammon Price to be important, you'll definitely want the most accurate MET you can get. -- Marty Storer I remember seeing experiments in which bot played a large number of matches aginst the same bot, same strength, only with one using an optimal MET and the other a semi-optimal MET. If I recall, optimal MET got a 50.1% - 49.9% edge on a longer match length. It does not matter much as long as you use one MET in one calculation and the MET makes sense (anything but Zadeh). -- Casper van der Tak Hosted by www.Geocities.ws