SSL Discussion Thread (#6: 19XX: The War Against Destiny)

[chum]

Nitpicking: you can win the first round, your opponent just doesn't have a timer that tells her to die. You have to use a character that can trap your opponent (Yumemi with the crosses, for example) or just get really lucky, basically.

[ebarrett]

where's the score thread, I have a score to submit

don't tell me heisenbaby is still tapdancing around putting up a scoreboard for the game picked by one of the winners and voted by like absolutely everyone

[trap15]

The condescension, it burns.

[Aleksei]

oh wow, this was probably inspired by TSS, its so fun, passed easy, normal and almost passed hard, I attempted lunatic and got beat on lvl 2... I've got a lot of practicing to do, or I could just play the new COD

[heisenbergman]

I was thinking that in the event that SSL continues on with relatively steady participation, the "league" can be broken up into "seasons", with each season being comprised of ten (10) games. Meaning, after 10 games, the scores for that season are final and everyone's score is reset to zero for the next season of SSL.

Also, just to maintain some variety in the game choices, let's try not repeating games in the same series within the same season. Meaning, for example, if Raiden Fighters 2 was chosen for this season, RF Jet can't be played until the following season. If DoDonPachi was already played, other DoDonPachis can't be played again until the next season. If a Touhou game has already been played, another one would have to wait until later. If Gradius was chosen previously, then... ... ...well, you get the idea.

Anyway, just that. As for PoDD, I'm looking at starting it on Wednesday probably... provided that all of the nominated games for the SSL after that are PM'ed to me by then. Will also send PMs to some other people who ranked high in SSL #2 to nominate game titles just in case I don't hear from those I originally PM'ed.

Cheers.

[Eno]

1- Shit is tough.
2- Sure, Lunatic is arcade difficulty, but touhou is harder.
3- Took a while to realize how chaining works, feel dumb now.
4- How the hell the opponent manages to shoot away 3 boss attacks in a row.
5- Marisa looks way more badass than in the Windows games.
6- Tough stuff.

[DocHauser]

Just noticed something with the league placing: ciox is one place ahead of me, despite only taking part in the 2nd week, where I finished ahead of him! You might assume that it's because my previous week's score is dragging my average down, but both my individual scores were higher than his one score.

He scored 22.52 in week 2.
I scored 23.74 in week 1 and 29.07 in week 2

Remind me again how this works?

[heisenbergman]

Yeah it's still a bit wonky due to the small sample size. Same could be said about trap15's placement.

Will illustrate in more detail when I get home. On the road ob the phone right now. Heh.

[Eno]

Just noticed something with the league placing: ciox is one place ahead of me, despite only taking part in the 2nd week, where I finished ahead of him! You might assume that it's because my previous week's score is dragging my average down, but both my individual scores were higher than his one score.

He scored 22.52 in week 2.
I scored 23.74 in week 1 and 29.07 in week 2

Remind me again how this works?

Not feeling like working out the math precisely, but I'd guess that it is either:
- Heisenberg took the m parameter to be half the games played so far(i.e. 1) for this estimate and thus people who played only one game get to be ranked, and as we play more games and m rises that side-effect will become less accentuated.
- It's what has been brought up on the organization thread, that just submitting a low score(compared to the current best score ofc) might harm your overall ranking rather than improve(or not change it).
Or both.

[heisenbergman]

Actually, I checked IMDB (that's where I got this particular Bayesian estimator from) to see what value of m they use, and apparently their current value is 25,000. Considering that their film with the most ratings is at roughly 1M votes (Shawshank Redemption), this means that their current m value is approx. 2.5% of the max ratings.

If I use that value of m for the SSL overall ranking formula, I get this table instead:

R=Rank & Player & SSLs Played & Average & WEIGHTED SCORE R=1 & Zil & 1 & 100.00 & 97.10 R=2 & Jaimers & 2 & 83.77 & 82.68 R=3 & Erppo & 2 & 81.36 & 80.34 R=4 & Zeron & 1 & 73.20 & 71.58 R=5 & ResumeDPosition & 1 & 62.35 & 61.25 R=6 & saucykobold & 1 & 59.32 & 58.36 R=7 & ratikal & 1 & 56.56 & 55.73 R=8 & chum & 2 & 55.42 & 55.02 R=9 & Zerst & 1 & 55.54 & 54.76 R=10 & Zengeku3 & 2 & 55.14 & 54.75 R=11 & trout & 1 & 51.56 & 50.97 R=12 & Battletoad & 1 & 45.40 & 45.10 R=13 & touhou=better.harder & 1 & 45.03 & 44.75 R=14 & Aleksei & 2 & 44.06 & 43.94 R=15 & Sakurei & 1 & 39.45 & 39.43 R=16 & jepjepjep & 1 & 37.30 & 37.39 R=17 & Lep & 1 & 34.16 & 34.40 R=18 & Aettaa & 1 & 31.68 & 32.03 R=19 & PAPER/ARTILLERY & 2 & 26.94 & 27.24 R=20 & VHS-porn & 1 & 26.43 & 27.04 R=21 & DocHauser & 2 & 26.41 & 26.72 R=22 & ciox & 1 & 22.52 & 23.32 R=23 & Chaos & 1 & 20.29 & 21.19 R=24 & Iori Branford & 1 & 18.17 & 19.17 R=25 & Eno & 2 & 17.77 & 18.29 R=26 & DMC & 1 & 17.03 & 18.09 R=27 & shmuppyLove & 1 & 15.67 & 16.79 R=28 & Pinehed & 2 & 15.64 & 16.21 R=29 & PedroMD & 2 & 13.81 & 14.43 R=30 & ED-057 & 1 & 11.39 & 12.72 R=31 & trap15 & 1 & 8.37 & 9.84 R=32 & heisenbergman & 2 & 6.74 & 7.53
So no more players with less games and lower averages ranked above players with more games and higher averages.

At the same time, there are also players with slightly lower averages ranked above players with higher averages by virtue of having played more games (see: chum over Zerst), the effect of which will become more pronounced as more games are played.

So that was my bad. The value of m I was using was way bigger that what I should have been using.

[Zengeku3]

I'm still not sure how your rating system makes sense. Take for example Ratikal being located higher than me, this is apparently because he beat me at Gunbird 2 but for DoDonPachi he didn't submit anything yet he's still rated higher than mine. I'm not sure this makes sense. Better yet Zil being on top of the ladder for being number 1 in one week despite Jaimers submitting good scores for both weeks.

In STGT you find the top standing players by giving everyone points relating to how well they did that week. Giving the player 0 points for not participating. In here it seems that if you submit a number 1 score one week then you are going to be the best until someone gets 2 number 1 scores? It's not like it's important for me who comes out on top or anything, for me it's more about competing within the weeks themselves but I'm just trying to make sense of your overall ranking system because it doesn't make sense to me that Zil can maintain the lead by just never diving below an average of 100 points unless perhaps someone else gets an average of 100 points having participated in two weeks.

[Eno]

Actually, I checked IMDB (that's where I got this particular Bayesian estimator from) to see what value of m they use, and apparently their current value is 25,000. Considering that their film with the most ratings is at roughly 1M votes (Shawshank Redemption), this means that their current m value is approx. 2.5% of the max ratings.

Sorry, but I don't get it. You said that you intend to use a 10 games then reset everything format, so if we go by that then at the end of a season we'll have that the minimum number of games that must be played to be ranked is floor(10*0.025) = 0. With that the ranking formula collapses to R and what Zengeku3 mentioned happens.
I find the old ratio to compute m to be much more reasonable, maybe even too low.

[heisenbergman]

You guys are right. I projected the scores to 10 games assuming Zil didn't submit a score again and he still ended up on top. Which doesn't seem right.

It's a bit tricky accounting for # of games played and how it should matter in the overall score.

Just to make my intentions clear, what I want to happen is to calculate a "weighted score" such that it takes into consideration that players who have played more games should have a higher weighting for their average than players who have played less games.

Take a Player A and Player B for example where the former has an average of 70 over 4 games and the latter has an average of 50 over 7 games... What I want to happen is that Player B should have a higher weighted rating than Player B because although his average is less by 0.06, he got it playing three more games.

Yet at the same time, I don't want the difference in games played to reflect overwhelmingly on the weighted rating. Take for example a Player W and Player X where the former has an average of 44 over 9 games while the latter has an average of 42 over 10 games. In this case, I think Player W having a slightly higher weighted rating than Player X is reasonable even though Player X played one more game than Player W.

I actually already took a look right now at math.stackexchange.com and I'm working out if the following formula seems reasonable:

Code: Select all

weighted score = average score * LOG(games played/total games*10)

I'm hoping that the logarithm of games played over total games will serve as a reliable-enough multiplier to account for the number of games a player participates in. If we use this, then this is the resulting ranking:

R=Rank & Player & Games Played & Average & WEIGHTED SCORE R=1 & Jaimers & 2 & 83.77 & 83.77 R=2 & Erppo & 2 & 81.36 & 81.36 R=3 & Zil & 1 & 100 & 69.90 R=4 & chum & 2 & 55.42 & 55.42 R=5 & Zengeku3 & 2 & 55.14 & 55.14 R=6 & Zeron & 1 & 73.2 & 51.16 R=7 & Aleksei & 2 & 44.06 & 44.06 R=8 & ResumeDPosition & 1 & 62.35 & 43.58 R=9 & saucykobold & 1 & 59.32 & 41.46 R=10 & ratikal & 1 & 56.56 & 39.53 R=11 & Zerst & 1 & 55.54 & 38.82 R=12 & trout & 1 & 51.56 & 36.04 R=13 & Battletoad & 1 & 45.4 & 31.73 R=14 & touhou=better.harder & 1 & 45.03 & 31.47 R=15 & Sakurei & 1 & 39.45 & 27.57 R=16 & PAPER/ARTILLERY & 2 & 26.94 & 26.94 R=17 & DocHauser & 2 & 26.41 & 26.41 R=18 & jepjepjep & 1 & 37.3 & 26.07 R=19 & Lep & 1 & 34.16 & 23.88 R=20 & Aettaa & 1 & 31.68 & 22.14 R=21 & VHS-porn & 1 & 26.43 & 18.47 R=22 & Eno & 2 & 17.77 & 17.77 R=23 & ciox & 1 & 22.52 & 15.74 R=24 & Pinehed & 2 & 15.64 & 15.64 R=25 & Chaos & 1 & 20.29 & 14.18 R=26 & PedroMD & 2 & 13.81 & 13.81 R=27 & Iori Branford & 1 & 18.17 & 12.70 R=28 & DMC & 1 & 17.03 & 11.90 R=29 & shmuppyLove & 1 & 15.67 & 10.95 R=30 & ED-057 & 1 & 11.39 & 7.96 R=31 & heisenbergman & 2 & 6.74 & 6.74 R=32 & trap15 & 1 & 8.37 & 5.85
So if this holds, I don't intend to impose a "minimum games to be ranked" anymore since players that have only played significantly few games after all 10 games have been played will naturally sink to the bottom of the rankings.

FYI, after 10 games, these will be the LOG multipliers depending on how many games one has played:

1 game = 0.00 (so if you submit only one score, you will have zero points after 10 games)
2 games = 0.30
3 games = 0.48
4 games = 0.60
5 games = 0.70
6 games = 0.78
7 games = 0.85
8 games = 0.90
9 games = 0.95
10 games = 1.00

I think this addresses the issue brought up in the organization thread where the ranking system should encourage more participation especially considering that there is less difference/variance in the multiplier with the more games one plays (i.e. - 3 games played vs 4 games played has a multiplier difference of .12 while 9 games played vs 10 games played has a multipler difference of only .05)

Thoughts?

[DocHauser]

I'm liking the look of that last system, mainly because I just jumped 5 places.

[Eno]

I like it. It's simpler, addresses the problems with the others and your weighed av. is always <= your av.(I'm not sure about the consequences, I just like it that way).
Not sure about other statistical implications though, not really good on statistics.

[Zil]

I might have at least posted a shitty 1-ALL for DDP if I'd known things would be changed around. w

Of course, I probably won't play anything after PoDD anyway so whatevs.

[DMC]

where's the score thread, I have a score to submit

don't tell me heisenbaby is still tapdancing around putting up a scoreboard for the game picked by one of the winners and voted by like absolutely everyone

Namecalling, really?
Your condescending attitude is tiresome to read.

[ebarrett]

where's the score thread, I have a score to submit

don't tell me heisenbaby is still tapdancing around putting up a scoreboard for the game picked by one of the winners and voted by like absolutely everyone

Namecalling, really?
Your condescending attitude is tiresome to read.

This entire forum is tiresome to read too, so I guess we're tied!

[DMC]

If the entire forum is at fault I dont think it's fair to take it out on one person, especially a new member that contributes to the hi score scene. Start a thread about the forum issue, or log off and take a break, rob a bank or play a game, all of which are more constructive approaches than these condescending oneliner posts, I believe.

[Aleksei]

rob a bank or play a game

Choices, choices, which should I choose today.

[CStarFlare]

Real men don't play games.

Just sayin'.

[KAI]

Real men are overrated.

[heisenbergman]

I'm liking the look of that last system, mainly because I just jumped 5 places.

I like it. It's simpler, addresses the problems with the others and your weighed av. is always <= your av.(I'm not sure about the consequences, I just like it that way).
Not sure about other statistical implications though, not really good on statistics.

Thanks for the feedback.

Not sure if I'm comfortable with how large the differences are between the LOG multiplers. Like... should an average of 70 over 4 games really be ranked lower than an average of 50 over 7 games?

I'm considering if maybe the proportions should be closer together so that the effect of playing more games is not as drastic as it is with that formula.

[chum]

I like the simple method of just assigning points according to placement (first place getting the same number of points as number of participants, last place getting 1 point)

The possibility of a poor score bringing down your rank shouldn't exist, imo, we should be encouraging people to play and try their best even at the games they are bad at. My 2 cents.

Also, for those playing the game, I broke the characters down into tiers in two categories (scoring and survival) hopefully it will be of use to someone if they don't know which character to play. I'll happily explain why certain characters are better or why some are worse, too, if anyone wants to know.

Survival:

Good: Yumemi, Reimu, Rikako
Average: Marisa, Kotohime, Mima
Poor: Kana, Ellen, Chiyuri

Scoring:

Good: Chiyuri, Kotohime, Mima
Average: Ellen, Kana, Marisa
Poor: Reimu, Yumemi, Rikako

The characters I would recommend are Reimu/Yumemi for survival, Chiyuri for scoring, Kotohime/Mima for balanced play. Keep in mind that more scoring = harder game, so use someone like Chiyuri at your own risk.

[Eno]

How do I bloody Mima? I've toned down to hard to train a bit but she's still ridiculously difficult. She's able to just spam boss attacks like no tomorrow and I'm luck to have a round where I can even get her to bomb, let alone hit her.

[DocHauser]

Just thinking of the scoring system again - at the minute, quite a few people are avoiding PoDD, possibly because they haven't played it before and don't think they'd score very highly. So, instead of a system that encourages people to play every game, it kind of forces people to tactically ignore games they're not good at.

So if you don't post a score, your score (SP) becomes zero for that game, e.g if ten people submit scores then everyone who doesn't is equal 11th (PP).

SP = 0
PP = 9.09
FINAL = 4.545

Obviously that's going to bring your average way down. It might be unfair to change the rules now, so it might be worth thinking about for future rounds...

[heisenbergman]

Thanks DH, that's definitely worth considering.

I'm thinking about it and I think I'm willing to stick with the latest incarnation of the final score computation for the meantime instead of assigning 0 SP automatically to non-participants. For starters, it seems like the pull downward might be too unfair; and secondly, I'd like to see how the current computation holds up to handling players that don't participate regularly.

[DocHauser]

For starters, it seems like the pull downward might be too unfair; and secondly, I'd like to see how the current computation holds up to handling players that don't participate regularly.

On the other hand, currently it seems like you'd be punished more for posting 9 good scores and one bad score than you would be for posting 9 good scores and one no-show. That doesn't seem right.

[heisenbergman]

Well, that depends.

If you post nine good scores, the one bad score would have to be less than your running average before opting out is a better option.

I'll keep it in mind, but would like to see how this current system works out.

[Aleksei]

damn it, I've been too busy with school to play for a good score and now I missed the deadline and get no score submitted