Subject: A new proposal on how to sort "Win Percentage"

Date: Tue Nov 27 09:58:23 2012
User: CubicSprock
Message:
While the +2 games works pretty well in the easy-to-medium difficulty scenarios, it falls well short of it's intended purpose (at least as interpreted by me), in the difficult variants. Take 4x6 for example, where the top 20 players in terms of win percentage are all 1 win in 1 game, after the win % mod, that puts them at 1/3 or 33%, still much higher than the percent of games that are actually winnable. Scrolling through the top of the list, I'd say #22 TNmountainman at 6 wins in 20 games is probably the most impressive since it has some statistical backing. Anyone can achieve 1/1 if they start a new nick for every game (tedious, but it's clear a number of people do it for just this purpose). Anyways, my suggestion would be to subtract 2 wins from every players instead of adding 2 games, while keeping games played the same...essentially count the 1st two wins as losses for purposes of sorting win %. So this would use (#Wins - 2) / #Games. So 1/1 would use -1/1...so in order to have a positive win percentage in a variant you would need at least 3 wins. This would have virtually no impact on easier variants and in tougher variants would give more meaning to win percentage by eliminating the 1/1 and rewarding consistent strong play over a larger number of games. thoughts?

Date: Tue Nov 27 10:41:38 2012
User: anicca
Message:
...you must be having great fun in there, CS... Eye don't even understand what you are saying~~ Keep up the work!

Date: Tue Nov 27 12:25:42 2012
User: TNmountainman
Message:
I've actually thought the same thing, CS, without ever proposing anything different. Well, actually, I thought of suggesting implementing a "minimum number of plays" idea, with that being somewhat variant *per* variant. Something on the order of 20-100 games played, depending on degree of difficulty of variant. Not many folks would really want to indulge in playing 100 of the super-difficult ones, but 20 (and not because that's the figure for me in 4x6 as you noted) seems a reasonable minimum. Or maybe 50? Any such change would sure re-sort the lists, and I agree that the change would more clearly reflect proficiency in that variant. To go back to baseball analogies (which I sometimes do), one can't win a batting title (etc.) without a certain number of at bats. On the other hand, many(?) of those players who have gone before, and 'retired' thinking their place in line was somewhat fixed, would not likely have the chance to come back and 're-secure' their standing. I don't think it's quite analogous to the winnable-vs.-regular debate, but I would be interested to hear Denny's opinion. If by some chance your idea was accepted and implemented, it would be wild to see previous leaders being changed to a negative winning percentage.

Date: Tue Nov 27 12:42:18 2012
User: jimmyp
Message:
How about adding wins AND losses to each person's totals? Add 50 games at a variant-average performance, so if 4x6 is, say, 10%, everyone starts out at 5 for 50, so the 1 for 1 becomes 6 for 51. Perhaps a little simpler, sort by "wins above average", so these 1 for 1 people would have been expected to have won 0.1 games and so they'd have 0.9 WAA. Hmm, unfortunately, this wouldn't work as well for the more winnable variants, as the leaders would just be the ones who played the most.

Date: Tue Nov 27 15:11:23 2012
User: CubicSprock
Message:
lots of interesting ideas, I think they are all getting at the same point...and I'm glad I'm not the only one who thinks there is an issue. The WAA is an interesting idea, but regardless of variant, it is going to rely too heavily on games played. For tough variants, say a 5% average win rate, someone who wins 10% in 10,000 games would have 1000 wins, 500 above average. Meanwhile someone at 20% over 1000 games would have a WAA of 150. This idea, similar to WAR in baseball, falls apart unless we set an upper limit to what constitutes a meaningful percentage. (think of a slightly above average baseball player playing in 1000 games while everyone else plays at most 162). In freecell you don't have that 162 game limit, so you regularly are comparing people with 10, 1000, and 10,000 wins...so the objective is to say at what point is win percentage meaningful, which is up for debate. Once a number is reached, you could normalize everyone over 'n' games, down to 'n' games and calculate win percentage based on that. Of course, at that point it'd just be easier to use a hard cutoff like TN suggests, and say you need 'n' games to qualify for the win % calculation. Anyway, I like the idea of adding wins and losses based on the average as that forces people w/o many plays towards the average. Theoretically it would be great, implementation-wise it would be tougher...how often are you recalculating that average? Are you using partial wins, what's the right number of games?...is this number variable based on variant difficulty. 50 games in 12x4 and you only add a fraction of a loss to everyone's totals, 50 games in 4x4 and you are only adding a fraction of a win to everyone's totals...which might be fine as long as computing fractional wins or losses isn't an issue. I think it's probably a little more accurate way that simply adding 2 losses (or turning 2 wins to losses as in my suggestion), but I'd also say it's less likely that Denny would implement it due to calculation of an average (and an ever changing one at that) that is variat specific.

Date: Tue Nov 27 18:43:24 2012
User: TitanicTony
Message:
Very interesting point, CS. Thanks! It certainly impacts my calculations of a "prolific winner" score (based on all 70 of ones %win ranks). However, since to get a good "Prolific" score, you need to play a lot (nearly all) of the variants, this anomaly might be minimized when scoring is done over 70 variants. Since the anomaly is only a problem for the most difficult variants, requiring a fixed minimum number of games played, a la TN, (say 10) for all variants would probably adequately solve the problem (of %win rankings). There might still be a few players who would use multiple nics (if enough players try often enough someone will eventually get to 3, 4, or 5 wins out of 10 (or maybe even 6, 7 or 8). Their choice, I'm not bothered. I have considered playing one 4x4 game, just to see if I could get lucky (but I think it will never happen). On a slightly different note, I am just slightly amused that you have to win a game to be ranked. I have NO IDEA how many players have actually played 4x4 -- maybe >1000? And, is 0/10 a better result than 0/20?

Date: Wed Nov 28 12:21:39 2012
User: CubicSprock
Message:
TT, I actually think the impact would be fairly significant with regard to prolific winner scores. You are correct that the people losing points would generally not have significant scores anyway as they are one game and done nicks. However, the people gaining points once those 1/1 players fell off the list would be players like TNmountainman (from my example) ...and it is precisely these types of players that are populating the top tier in the prolific winners rankings.