Subject: To GGG, on the 12-sum rankings

Date: Thu Apr 6 22:12:33 2017
User: joeygray
Message:
Hi GGG, I noticed your post today as I was checking my email/messages during intermission at a show I was attending, out of town. I typed a rather long reply but sadly the curtain went up, and when I turned my phone back on it was gone; I think I'm not gonna try any more lengthy emails from the phone. Anyway, let's try this again. First, tonight's auto run will just happen with the current calculation. If we change something it won't take effect til next Monday's run. Second, I've referenced the original thread that explains the calculation. Basically, the behavior that the sum of your wins being higher than than anybody's isn't resulting in your score being higher, was always the intention of this metric I designed, which was to measure the entire range of 12-sums. The higher your streak is in a variant, the less one more win in that variant will contribute. That was what our friend free-at-last came up with when he multiplied the square roots of streaks for his rankings. For this metric I did 'even more so' by using log-to-the-base-2, which rises considerably slower than sqrt does. I guess I could go back to sqrt but either way the idea is that one should lead this list by attacking all the 12-sums, not by piling up wins in the easiest of them. Third is your particular objection to 4x8. On the basis that you just don't like them, I'd be inclined to reject any changes. Cause after all it is a 12 sum, and I for one DO like them, I think they are quite challenging. You are better than than I am at them, so you know this. However, there is certainly another point, that they are 'different' than the other 12-sums. You expect to lose when you play one. You in particular hate losing and reject playing some variant where the expectation is losing, but even for me the tendency is to play them until I win one, then drop em entirely for a while. Like 14 days, maybe. One fix is to eliminate the penalty, which currently is to halve your score for any variant you don't have a current streak in. I could slightly change my software to eliminate the penalty for 4x8. And for any variant whose level-5 expectation is loss, as given by the win percentage of the best frequent players thereof. Like 4x7 for sure and maybe 5x6, I haven't looked. But note that this scores any zero streak the same as a one streak; you would still be handicapped by not playing 4x8, w/r to anybody who has a streak of 2 or more. Yours and anyone else's thoughts are welcome.

Link: Original post

Date: Fri Apr 7 03:44:25 2017
User: bozkurt
Message:
I think log2 is better than sqrt. Your penalty is something like log2(sqrt(2))=0.5 sqrt(2) is about 1.414. Instead of completely eliminating penalty, log2(1.75+streak) or even log2(1.90+streak) may be OK with a lesser penalty. if streak=0 then log2(1.75)=0.81 or log2(1.90)=0.96

Date: Fri Apr 7 05:38:27 2017
User: Goosey_Goosey_Gander
Message:
I like the logic of using square root or log so that the benefit of long streaks gradually diminishes and there is a larger reward for achieving a modest streak in the more difficult variants, but I don't see the logic of halving the total score for each variant without a current streak, it seems like a disproportionate punishment. Punishment for what? If two variants just happen to hit the 15 day window before you calculate a score it means the score is reduced to 1/4. I don't think that truly reflects current standing. I would vote for simply deleting the one half thing, any variant.

Date: Fri Apr 7 08:37:02 2017
User: joeygray
Message:
G, I think I agree, the halving thing was ill advised. I actually did it, as I said in the earlier thread, because it's better than zero and because I wanted 1 win to be 1 point and because I wanted one win to be better than no wins. So how about if I just either use log (x+2) or sqrt(x+1), removing the discontinuity entirely. 1 win won't be 1 point anymore but who cares really, it affects everybody's score the same. This isn't fixing your real problem about the 15 days, though. If you have a streak that you really mean to keep playing but Denny drops you off before you get around to it and I happen to run during that, boom, big loss of points. Of course I *could* go look up your actual streak on your individual page, I already have code that does that which I used to run the all time stats, but I can't tell your intent: are you really 'intending' to continue this streak or (as in 4x8) are you intending not to. So I'll stick with Denny's arbitrary decision, and only count streaks he is counting as current.

Date: Fri Apr 7 09:50:39 2017
User: joeygray
Message:
Ok, so this is the results of last night's run using log2(x+2) instead of if(x,log(x+1),.5). Look on the other page to see the old one.

	4x8	5x7	6x6	7x5	8x4	9x3	10x2	11x1	12x0	count	Score	As of 4/6/2017 11:01:49 PM
==	===	===	===	===	===	===	===	===	===
1	4	11	35	55	230	110	75	45	14	9	2,164,859	bozkurt
2		7	10	75	125	89	458	65	37	8	918,579	Goosey_Goosey_Gander
3	5	7	25	35	35	31	50	23	15	9	626,788	Zonny
4	2	6	12	28	62	240	56	18	14	9	539,362	Kaos
5	3	20	21	13	64	102	10	38	11	9	523,243	joeygray
6	1	9	12	14	10	14	15	12	5	9	52,315	jimmyp
7	1	2	4	7	54	11	8	8	8	9	20,462	Lindyhopper_Agame
8	1	2	2	5	21	19	35	19	3	9	18,788	FairyGrandma
9				7	193	14	13	56	7	6	6,998	Punster
10	3	10	20	57	17			17		6	3,940	ElGuapo
11		6		10	5	80	25	10		6	3,272	Mastermind
12				26	2	17	16	21	14	6	3,082	ListeningtoGod
13					118	443	40	52		4	1,886	T1-T3
14		2	45	5		10		20	9	6	1,725	gilbertgame27171
15	1	4			19			37	19	5	417.77	TNmountainmanMUDstains
16			4	34				150	3	4	224.91	gilbert10021
17					22	2	11	1	3	5	124.88	jsj41
18					11	88	2	1	1	5	120.70	d164280
19		3	2	9	25					4	76.39	RogerAnonymous
20		1		3		13			2	4	28.76	ValpoJim
21			1		3	6			2	4	22.08	larwul

And this is is the result using sqrt(x+1)

	4x8	5x7	6x6	7x5	8x4	9x3	10x2	11x1	12x0	count	Score	As of 4/6/2017 11:01:49 PM
==	===	===	===	===	===	===	===	===	===
1	4	11	35	55	230	110	75	45	14	9	12,753,216	bozkurt
2		7	10	75	125	89	458	65	37	8	9,343,827	Goosey_Goosey_Gander
3	2	6	12	28	62	240	56	18	14	9	1,397,394	Kaos
4	5	7	25	35	35	31	50	23	15	9	1,006,782	Zonny
5	3	20	21	13	64	102	10	38	11	9	944,299	joeygray
6	1	9	12	14	10	14	15	12	5	9	28,339	jimmyp
7				7	193	14	13	56	7	6	12,191	Punster
8	1	2	2	5	21	19	35	19	3	9	11,699	FairyGrandma
9	1	2	4	7	54	11	8	8	8	9	10,746	Lindyhopper_Agame
10					118	443	40	52		4	10,715	T1-T3
11	3	10	20	57	17			17		6	4,167	ElGuapo
12		6		10	5	80	25	10		6	3,271	Mastermind
13				26	2	17	16	21	14	6	2,860	ListeningtoGod
14		2	45	5		10		20	9	6	1,383	gilbertgame27171
15	1	4			19			37	19	5	389.87	TNmountainmanMUDstains
16			4	34				150	3	4	325.12	gilbert10021
17					11	88	2	1	1	5	113.21	d164280
18					22	2	11	1	3	5	81.39	jsj41
19		3	2	9	25					4	55.86	RogerAnonymous
20		1		3		13			2	4	18.33	ValpoJim
21			1		3	6			2	4	12.96	larwul

Date: Fri Apr 7 09:59:17 2017
User: joeygray
Message:
Yeah, so you can see that sqrt changes some positions a little, as log2 gives greater weight to spreading out your efforts over all the variants... so log2 it is! :). Starting with next Monday's run.

Date: Fri Apr 7 10:57:56 2017
User: Goosey_Goosey_Gander
Message:
Good job jg, thanks. I agree with everything. I don't have a problem with the 15 day thing, I just have to make a point of playing a game after 14 days instead of after 15. I may even play 4x8 and 4x7 once in a while. Arrrgggg. Now to chase bozkurt.

Date: Fri Apr 7 11:16:43 2017
User: joeygray
Message:
Sigh. On the other hand, I'm thinking about the disproportionate effect of 4x8 on this metric. Obviously, the next game to play if you want the maximum effect of a win is the variant you have the lowest streak in. That's always gonna be 4x8. But chances are, you are going to lose that game. It just seems as if the leader is going to tend to be the guy who can stomach playing a flurry of these thing every 14 days or so until he gets to 4 or 5 long streak. I certainly never wanted to make 4x8 MORE important than any other variant... so what do I do?

Date: Fri Apr 7 11:49:54 2017
User: TNmountainman
Message:
Obviously (.grin), instruct all players to play the climate 6 games and tell the scorekeeping system to have them count in streak play. We all know the climate 6 games are easier, right?