New 3bv Limits/New rule for lucky boards?

[Tommy]

I'm starting this thread although this is actually Ronnys idea.

Ronny wrote an excellent article regarding the rareness of boards and the effectiveness of 3bv limits after Ians 33, and I propose we discuss his conclusion here.

For lazy readers, in the article 3BVO is defined as 3BV+2*(openings-1) to account for the fact that openings usually take more than one click to solve (except for the first one).

After statistical analysis, Ronny reaches the conclusion that expert boards with 3BV<100 are much rarer than intermediate boards with 3BV<30, and that the current 3BV limit on expert is not effective. This is because the 3BV distribution of expert highscores on the world ranking is hardly affected by the 3BV limit at all, and most expert highscore boards have a 3BV that is far above the limit, in contrast to intermediate, where most highscores have a 3BV that is close to the limit.

You really should read the article though: http://www.dmc-hq.nl/west_images/minesw ... _trial.pdf. Especially the last page or two.

My personal opinion is that we might want to consider ZiNi (http://minesweeper.info/forum/viewtopic.php?f=15&t=70) and do the same kind of analysis for that stat as well.
Apart from that I really do think that we need to change the limits, be 3BVO or 3BV (or something else) the stat we use.


That all said, once again, great job Ronny

[EWQMinesweeper]

i think that zini/hzini say a lot more than 3bvo.

in almost 1 month of sweeping with v47 arbiters, which feature zini counters, i've gotten 2 or 3 boards with zini<100 and tons of boards with hzini<105. ian got his 33 @ ~2.69 hzini/s, my exp record on 162 3bv, 184 3bvo, had only 115 hzini (109 zini) @ 2.77 hzini/s...


my suggestion: 30 and 90 as hzini minumum limits

30 because everything lower on int would be too dreamboardish
90 because ian's 33 had 89 i think (it should still remain accepted) and sub100 hzini doesn't seem that rare to me


my opinion: do not dare to touch the current 3bv limits it's that hope for such a lucky board like ian's that keeps us motivated to still play equal chances for everyone! if ian is allowed to get such an easy board, i won't accept them being banned in the future, making it impossible for me to get equally easy boards


happy sweeping!

[RonnyDeWinter]

@[name removed]: Personally I think ZiNi or HumanZini is to complex to be able to introduce it successfully. I've also run all boards with ZiNi (I don't know about HumanZini) and the results aren't all that different than 3BVO, because the same boards end up in front as well.
About not touching the 3BV limits, because you like having a chance of beating people that are better than you is something I personally can't comprehend. What's the fun of beating someone else or yourself on luck instead of skill? From what I've seen is that players that got a lucky record is that they ended up demotivated, frustrated, a bit ashamed and/or had very mixed feelings about their records. It may be cool for a day to get an insanely lucky record, but after that the glory fades away quickly. It's the same with other sports, like running, you don't run downhill to get a personal best or run with a hurricane wind blowing in your back, but you try to improve to get a personal record. Also I think it's a bit weird that you feel a lucky record for you on expert should always be allowed, while you think that a Kamil's 3BV=26 flag record from yesterday shouldn't be. That would be measuring with two different standards in my point of view considering that his 3BV=26 is 25 times less rare (3BVO) than Ian's record.

I don't know enough about human ZiNi yet to judge its effectiveness, but if you guys want to know more about what would be equally rare ZiNi limits based on the current intermediate limit, I will run the same tests with HumanZini and check and post the results later.

3BV: 30 vs 109/110
3BVO: 36 vs 134
hZiNi: ? vs ?

[KamilSaper]

I believe that there should be 2 rankings - 1st without limits that Ian's 33 would be accepted, 2nd with increased limits on exp, so probably Ian's record woulad have to be removed

[RonnyDeWinter]

I've just noticed that HumanZiNi might not be not such a bad idea if you would only change the initial ZiNi value from 'ZiNi = [number of openings]' to ZiNi = [number of openings]*3 - 2. This way you would merge HumanZini with 3BVO.

I'll see if I can run both test, being with the current HumanZini and the merged 3BVO/HumanZiNi to see what results in the best results. Also I will add the suggestion of Bertie that I should also analyze the effect of a limit of 3BV=109 or 110.

[Cryslon]

I think H.ZiNi is pretty good benchmark, eg, dreamboards' ZiNi value (dreamboard thread) is much lower than 3bv (which meets the limit) of these boards. H.ZiNi isn't that difficult to implement and it's pretty easy to understand (though i don't think that for good measure it's required to be simple). The main problem is absence of one of the clonemakers...

I also strongly dislike ideas about splitting the rankings. I think there should be only one time ranking, one #1 and one record holder for each level. Also, rankings maintainer doesn't have infinite amount of time.

And Ronny, why don't you join IRC?! I guess there are many people (at least me) who want to talk to you live.

[RonnyDeWinter]

Actually there already are 2 rankings being the World ranking and the Decimal World ranking. The decimal ranking already has more strict rules and is already more fare by summing decimals instead of looking at the expert integer value. However, I personally also am in favor of increasing the limit on expert in the common world ranking as well to make it equal to the limit on intermediate.

Yesterday I've been busy with HumanZini as well, at first notice it does seem to be a better indicator than 3BVO, but I see some contradicting results as well which I don't understand yet. I had this assumption that the best indicator would show the biggest influence on the normal distribution of random boards when comparing it with the world ranking boards, but so far HumanZini and 3BVO came out worse than 3BV and I'm doubting my method of comparing, so I'll try to verify or disproof that again tonight.

[Cryslon]

Yesterday I've been busy with HumanZini as well.

I'll try to use this place as chat (i hope [name removed]and Damien won't ban me).

I'm interested in a way you make your research. Have you upgraded zinicalc to make it easy to get info about large number of boards? Do you consider only record boards from ranking?

@3bv limits: do you want to raise them? And you wanna drop Ian's 33? I don't think it's good solution. Also, how would you set the new limit? 110 is still too low imo comparing to 30...

[KamilSaper]

I don't want to drop Ian's 33, so I'd like to dicrease the limit on int As I've written in the GB, sub30 boards occur too often to be considered as too easy. It would also solve the problem of some accepted sub30 boards it the world ranking.

[Tommy]

Now way @ reducing the int limit. It works well and enables people to get highscores without playing hours just to get a board that is good enough. That doesn't measure skill as much as it gives a big advantage to players that play all day.

30 cuts the left branch off the highscore bell curve extremely nicely. And if we create new 3bv limits, I'm guessing that we don't need to apply the new 3bv limits to old highscores. It's not like nobody is ever gonna beat that 33.
sub50 seemed impossible for a while. Until a couple of months ago, the same was true for far sub40 (I remember when the best players in the world had 39, and that seemed to push the limit).
Matt McGinley and his 10 on the dreamboard are a similar case. Nobody worries about that 10 still being on the ranking although it would certainly not be acceptable now, and I guess that we can say the same if we create new 3bv limits. Heck, it was an ex aequo world record until recently. Everyone knows the background, and I guess that we can live with old scores that wouldn't be let into the ranking knowadays if it makes minesweeper a better game.

EDIT: The above assumes 3BV limits, of course. It applies to a change from 3bv to 3bvo on int as well though. We can have a couple of old highscores that don't fit in with new 3bvo limits, especially as the changes ronny proposed only make a very slight difference.

[EWQMinesweeper]

@max: feel free to post as much as you want

@sub30 int boards being not so rare etc: i never got one on winmine (and at that time i was playing a lot on all levels + i never got one on msx, though i haven't played much int on it). i for my part get sub130 3Bv boards on exp at least once every hour - from my point of view there are enough easy exp boards well above the limit. i do not see any use in raising the exp limit just because of a single very easy board. if you increase the limit now, you would have to declare ian's 33s board as too easy and unacceptable for the rankings, otherwise, since it'd become much harder to get such a score on a board fitting the new limit, things would become a bit unfair to sweepers who haven't had the chance to get highscores with the current 3bv limits being in use. do you really want to drop ian's 33?

so why don't we go for a hzini limit on exp instead? if we chose eg sth ~90, people would still have the opportunity to get boards as easy as ian's, but it would erase the possibility of such big record jumps as ian's 39->33 drop.

[RonnyDeWinter]

Yesterday I've been busy with HumanZini as well.

I'll try to use this place as chat (i hope [name removed]and Damien won't ban me).

I'm interested in a way you make your research. Have you upgraded zinicalc to make it easy to get info about large number of boards? Do you consider only record boards from ranking?

@3bv limits: do you want to raise them? And you wanna drop Ian's 33? I don't think it's good solution. Also, how would you set the new limit? 110 is still too low imo comparing to 30...

I haven't changed zinicalc a lot really. I only added the possibility to create random boards and to get the normal distributions of 3BV (3BVO, hZini or any other algorithms) I've just created millions of those boards and counted the occurrences of the results. After that I output these results in a comma seperated file and do all the rest of my calculations in excel. For the graphs of the world ranking scores I basically do the same but instead of using random boards as input I use a folder containing all the board/video files instead. After that I once again to all the other calculations in excel.
World records boards were a bit tricky because those were mainly in unreadable formats like jpg and avi, so I had to create those board files manually.

About the 3BV limits, yes I do think it's smart to raise the 3BV limits of expert or otherwise add an additional limit to rule out the easiest boards like we already do on intermediate, so both will be equally fair. What we should do with Ian's record.....I don't know for sure, because both removing it and letting it stay in world ranking feels equally bad. If you let it stay in it will be hard to fix the incorrect 3BV limits on expert, if you remove it you punish Ian for a mistake that the IMC made in the past. Personally I hope that Ian keeps improving to either beat/tie this lucky record himself or that Kamil or someone else beats it soon, because that would make it 10x easier to fix this mistake of the past.

Right now I'm focusing more on supplying the statistics, rather than what we should do with it. At the moment I still haven't got proof of which limit and algorithm is best. 3BVO and hZiNi both look nice in theory and seem to work nicely when applied on easy boards of the past, but I'm still haven't found real proof that it's truly any better than 3BV. I only can proof that with any algorithm you apply, that the current expert limit is chosen very very poorly.

[EWQMinesweeper]

About the 3BV limits, yes I do think it's smart to raise the 3BV limits of expert or otherwise add an additional limit to rule out the easiest boards like we already do on intermediate, so both will be equally fair. What we should do with Ian's record.....I don't know for sure, because both removing it and letting it stay in world ranking feels equally bad. If you let it stay in it will be hard to fix the incorrect 3BV limits on expert, if you remove it you punish Ian for a mistake that the IMC made in the past. Personally I hope that Ian keeps improving to either beat/tie this lucky record himself or that Kamil or someone else beats it soon, because that would make it 10x easier to fix this mistake of the past.

how would someone getting an better score make things easier? ian's 33 then still stands there....as far as i know ian hasn't been in the irc-chat a lot since his semester at uni started again -> he probably doesn't have nearly as much time to sweep as when he got his 33etc.

so what about declaring ian's 33s board the example of an exp dreamboard and for the future banning all easier boards?

[Cryslon]

... to get the normal distributions of 3BV ...

What do you mean by 'normal distribution' ? Looks like 3BV distribution (at least on int and exp) isn't similar to gaussian, right wing obviously falls slower than left.

[Tommy]

Not much of a difference though

I don't think we need to argue exactly what the distribution looks like, we all know and understand the implications.

The question is what to do know.

The point I was trying to make with my last post was that I think that we overestimate the effect of leaving the 33 in the rankings while raising the limit for future games.
Basically this situation seems much like the situation with Matt McGinleys 10 - and I think that we can raise 3bv limits (ban the dreamboard) and still leave ians 33 (matts 10) in the ranking.

[Cryslon]

The question is what to do now.

Well, and what's our aim?

If we're going to ban extremely easy boards, i think HZiNi limits is a nice solution, at least until somebody brings dreamboard meeting our limits.

Ian has already scored the 33, so it's imho too late to raise 3bv limits. You can't just go and drop it, saying 'we've decided to introduce new limits'. And if you drop the 33, what would you do with Dan Cerveny's record scored on a 99 3bv? Also, even much higher limits, eg 140, still allow dreamboards, see dreamboard thread.

About consequences of leaving the 33: i can't say much because i can't even score a sub50. But 120 3bv board with 5 cl/s and 0.8 ioe (possible stats, aren't they?) gives 30, so i think 33 is theoretically beatable.

Also, this discussion is very similar to usual old, long, leading-to-nowhere GB ones. The men who decide are Damien as ranking maker, and clonemakers who add the limits to programs we use. I fear we may argue here infinitely long while no decision is made.

In addition i want to say that i dislike the idea 'leave it as it is'. Ian's board is very rare, but current limits are met by even more rare and lucky boards, we should ban them NOW, IMMEDIATELY, before anybody scored sub20 or sth about on them. And do not tell me that these lucky boards are extremely rare, we should exclude this possible situation completely and shouldn't rely on chance.

[RonnyDeWinter]

The point I was trying to make with my last post was that I think that we overestimate the effect of leaving the 33 in the rankings while raising the limit for future games.
Basically this situation seems much like the situation with Matt McGinleys 10 - and I think that we can raise 3bv limits (ban the dreamboard) and still leave ians 33 (matts 10) in the ranking.

Probably right. It's not like the 33 is likely to be beaten with a just as rare board. Increasing the limit is more to make the world ranking and WR list more fair on both levels in the years to come.

[RonnyDeWinter]

Right now I'm focusing more on supplying the statistics, rather than what we should do with it. At the moment I still haven't got proof of which limit and algorithm is best. 3BVO and hZiNi both look nice in theory and seem to work nicely when applied on easy boards of the past, but I'm still haven't found real proof that it's truly any better than 3BV. I only can proof that with any algorithm you apply, that the current expert limit is chosen very very poorly.

I took me a few days to figure out a way to fairly compare 3BV, 3BVO and hZini (or a combination) to find out which method most fairly compares boards on easiness. Yesterday I finally had a brain wave that resulted in a nice comparison method. Only downside is that probably now have to write an article '3BVO and HumanZini on trial', because it seems some of my conclusions about its effectiveness were apparently incorrect and I also now have statical evidence about how good or bad HumanZini works in comparison to 3BVO and 3BV.

Anyway, I haven't put all result on paper yet, but the first results are quite interesting and will probably tell us more about whether sticking to 3BV or changing to another method is best and which limits are best to use on expert and intermediate.

I'll post the results within 1 to 3 day I think, so Cryslon will have to wait a little longer before he knows whether HumanZini kicks ass or was only nice in theory, but worse than 3BV in reality.

[Tommy]

Come on, you can't be serious that we aren't gonna do anything.

Even though 3BVO and ZiNi are far from perfect, I guess we can agree that the current limit is far from effective and needs to be changed.

Minesweeper isn't fun when your only chance to break your or someone elses highscore is to play for hours every day so that you might get a great board and finish it. I don't have the time for that, and this community has always mainly consisted of players that do other things in their free time as well.

Luck is needed to get a highscore already. The more rare the best boards are, the more getting a highscore depends on luck. The ranking should try to represent skill which might mean a number of things, but certainly not luck.

[RonnyDeWinter]

Don't worry. This little pause in the discussion gives me the required time to finish my experiments to see which indicator can be used best. I've already finished all testings and calculations, but I still need to add the final chapter and conclusions to the report. It should be ready this weekend.

[Tommy]

cool

Especially as this means that I can read it before doing military service I probably won't be online for two weeks starting from monday.

[Cryslon]

Come on, you can't be serious that we aren't gonna do anything.

Hehe, more than month passed and we've done nothing, eh? That's pretty usual situation, isn't it? Almost all active members state that current rules are crap, but nobody makes new rules, because there are no responsible persons.

BTW, does anybody communicate with Rodrigo?

[KamilSaper]

I told him he enter IRC channel, but probably he's not going to

[RonnyDeWinter]

Sorry for not making it before monday. I just installed Windows7 which took some hours and needed to get another Office before I could continue (I was using Office 2000 till now). Anyway, I only have to do a spelling check and finish the conclusions, so hopefully tomorrow.

[Cryslon]

Use OOo!