Official: SSBPD unsupported; source code released.

[Fortress | Sveet]

Anyways I'm off tomorrow so I'm gonna play some SC2 tonight

with me?

 

[Kal]

omfg stay on topic sveet come on ur in the mbr u should no better

 

[Fortress | Sveet]

On a related note, I knew it was going to rain today. I washed my car yesterday.

 

[Fizzi]

This is just an observation but it seems at present, the website requires a relatively large amount of moderation by admins. I feel like this is a weak point of the website as it would be hard to maintain for a long time as people get bored/lazy. It would be a more sustainable website, overall, if the means were added for it to be community operated.

In a previous post I mentioned region groups/subgroups. This would be cool not only for rankings within subgroups but also a person uploading a tournament could specify which primary group it pertains to and then there could be 2-3 trusted members from that group which can go through and confirm each player.

For example, a tournament hosted in NorCal but gets attention from all of WC US would be uploaded as a WC US tournament. Then trusted members from WC US (Which were likely at the tournament) could confirm each player is being submitted to the correct entry and pass the tournament themselves. Note that it would be possible to someone to be a moderator for multiple groups, for example (WC US, Cali, NorCal - if the person lives in NorCal but is reputed throughout WC).

 

[KishPrime]

I thought I already forwarded the PM i received from AZ. JV might be able to, though. Would be best if we could get archives from the archived SWF threads.

Jv might be able to get them, but he wasn't running it then. That was M3D. I don't know that brackets were really kept for the events back then, though. I have a bunch of match results from MLG Chicago 2004, but not brackets.

 

[Divinokage]

It's still not a quick as a process as you would think. It probably takes about 10 minutes to process a 100 person tournament. All I really do though is go through the entrants and make sure their name in the file matches what's in the DB because it's easier to fix beforehand than after. Also I saw your list and actually made a much larger one of "target" tournaments to have in the DB. It's posted a few pages back and has over 75 tournaments in it dating back to 2007.

Anyways I'm off tomorrow so I'm gonna play some SC2 tonight and then hit the DB hard with the goal to process every tournament that is currently uploaded, resolve all player flags, and upload at least 10 tournaments from our list. Should be fun

Ahh I was about to ask how long it actually takes to process one tournament. Well now I know, =P,

 

[Deleted member]

lol mine is soooooo wrong.

cool idea though. i like it.

 

[Divinokage]

I am confused about something. Why do I have the Pound 3 data and Kirbykaze does not have the Pound 3 data for himself? Isn't it suppose to handle all players that attended?

 

[Scar]

I went ahead and created a form for a bug reporting tool. Use this to help us collect bugs!

 

[ajp_anton]

Afaik, ELO is based on how often player A beats player B. However, this ratio varies between different "best of"'s.
If the match ratio between A and B is 80%, then in bo3 and bo5 the ratio is 90% and 94%.
Or with a match ratio of 90%, bo3 and bo5 are 97% and 99%.
This could result in a difference of hundreds of ELO points.

This doesn't matter much with even players, but even if one could take 10% of matches against Armada, it's practically impossible to win a single best of 5 set.

I know there isn't much that can be done about it, but it's still a flaw in the system that it mixes sets of unknown length, and doesn't know about 3-0 vs 3-2 match ratios.

 

[Kal]

You could implement the Elo to calculate based on individual wins, but you can also calculate based on who wins the set (which is how I believe it is calculated for SSBPD).

Or did you mean rather that it's unfair that the Elo does not distinguish between a 3-0 and a 3-2 victory?

 

[ajp_anton]

My point was the difference between bo3 and bo5 sets. And over here we often do bo7 in the finals.

The 3-0 vs 3-2 was just another example of inaccuracy. The problem is that with so long sets, the win/loss ratios get much closer to 1/0 and you need 10 times more results to actually get some variation.

If you use for example bo3 as the basis, I think you could "guess" the set length for different bracket matches, and weigh their results based on that. And in pools you should have the results for individual matches, which again would be weighed differently.

 

[Kal]

Well, I would say that the win-loss ratio in a set is irrelevant. In a tournament, the only thing that gets registered (in bracket, at least, but usually in pools unless you are breaking a tie) is the result of the set: a player does not go further if he wins 3-0 instead of 3-2. Thus, I think the ranking should factor in only what the tournament itself factors in: the results of sets. To me, it would be ***-backwards to have a tournament in which "winning 3-2" and "winning 3-0" are identical results, but to have rankings based on said tournament in which "winning 3-2" is worse than "winning 3-0."

However, a point that you're getting at is that different tournaments use different rules. A tournament with best of three is, in a very real sense, a different game than one with best of five or best of seven. More generally, it's an issue that we have different stage lists and such available at different tournaments. In other words, we're all playing different (albeit incredibly similar) games, and compiling results from all of these different games to determine who is the best. The only real fix to the problem you have described is to only use tournaments which follow a specific ruleset.

However, I think the issue is minor. In general, if a player has a very high Elo, he will perform well regardless of the rules. Similarly, if a player has a very low Elo, he will perform poorly regardless of the rules.

 

[Scar]

My point was the difference between bo3 and bo5 sets. And over here we often do bo7 in the finals.

The 3-0 vs 3-2 was just another example of inaccuracy. The problem is that with so long sets, the win/loss ratios get much closer to 1/0 and you need 10 times more results to actually get some variation.

If you use for example bo3 as the basis, I think you could "guess" the set length for different bracket matches, and weigh their results based on that. And in pools you should have the results for individual matches, which again would be weighed differently.

in bracket, matches don't matter - a player who wins every set 2-1 or 3-2 or 4-3 is still the best. i think the argument is better made re: pools, which, coincidentally, is another example of how swiss is a more elegant seeding tool than pools.

 

[Bones0]

Yeah, matches shouldn't matter. For example, you may have some players who do worse game 2 because they know their opponent's CP is too strong, and instead they choose to focus on gathering a bunch of patterns that they can abuse in game 3 on their own CP.

 

[FrootLoop]

some tourney's pools have 2-0's recorded as 3-0's. An example is smym13. 2-1's are still 2-1's though.
I guess it doesn't matter, but just fyi.

 

[ajp_anton]

How can anyone possibly think "matches don't matter" and "if you win, you're simply better"?

The only thing ELO is designed to measure is how often a player beats another. If the "better" player always wins, ELO has no meaning. It's all statistics, and having different number of matches per set alters the statistics.

If we want it to measure how often player A beats B in a best of 3, then we need to alter the calculations for results that weren't best of 3. Taking into account the actual match score isn't necessary, but it would make it more accurate with fewer encounters.
I don't know how it currently calculates pools, where the match score is automatically available...

Anyway, I guess it still works to a certain degree of accuracy, so if you think it's good enough, just ignore this.

 

[FoxLisk]

A set is a unit of victory or defeat end of story.

 

[Fortress | Sveet]

The player that plays better wins, but thats not to say that being "better" is set in stone. Elo is a mathematical way of keeping track of who is generally better in a relative way. There are way too many factors to ever say who is better with certainty.

 

[Scar]

yeah +10 to sveet and foxllisk.

don't forget that tournaments aren't necessarily indicative of anything, but we use them as a tool to measure who is the best player on that day. and as far as the tournament is concerned (excepting pools), it gauges "best player" solely based on set victories.

 

[Deleted member]

why is kels #4 elo ****ing midwest

 

[Fortress | Sveet]

Cause kels is a beast.

 

[Zivilyn Bane]

AJP, Sveet and I had a similar discussion in the backroom as to why we couldn't (and shouldn't). There is a simple fact that NO MATTER WHAT, a player should not lose points for winning a set and nobody should gain points for losing. If a player rated 1200 loses to a player rated 1600 by a score of 1-2, this can happen. Both players K value is 32, which is what 1600 would lose for the one loss. But that same player would only gain about 4 points for the two wins (you don't gain much for beating much lower rated players). That would net -28 points for a guy who won a set, which cannot be allowed. There are other reasons as well but we don't need to go over them because that first reason is all you need to understand.

EDIT: S2J, when we recalculate he will drop quite a bit, probably 75 points. When we have enough data everything will be a lot more accurate.

 

[Nintendude]

lol mine is soooooo wrong.

cool idea though. i like it.

The system is only as intelligent as the data it has. Go upload tournaments from back in your day if you want to see your rating increase in accuracy.

 

[Varist]

people be fake as **** thinkin kels isn't better than m2k lucky pp shroomed amsah and wobbles combined

i mean come on he's only ever lost to
every good player that ever went to a tournament
at least once so why shouldn't he be number 4.


but seriously i hope kels is still in the top 10 after elo degradation and stuff is put in and stuff and whatever it does.

because that would be hilarious

dat boy grindin out matches against randoms like whaaat

 

[Nintendude]

As soon as people upload more large tournaments that Kels actually went to, where he loses to top players, his rating will fall into the range it should be in.

 

[Fortress | Sveet]

Or people will realize that he actually has a lot of legit wins and deserves his spot

 

[Zivilyn Bane]

Or people will realize that he actually has a lot of legit wins and deserves his spot

Kels is amazing and likely deserves at least an 1800 rating, but he is definitely not 4th in the world.

I think what people are failing to realize is that the idea goal is to have people like Armada, Mango, Hbox, M2K, and other elite players in the 2200+ range, and great players like S2J, Shroomed, Kels, ect in the 1900-2100 range. When we have enough data it will much more closely represent the skill levels of the players though.

EDIT: And Varist, please keep your trolling spam out of this topic. You're the only person in here trying to act cool. Everybody else is talking about how amazing this project is and how it will benefit our community.

EDIT2: Whoever uploaded Smash T'ill You Crash 2 - it has the events but no matches entered, no pools, and a blank bracket full of bye's.

 

[Fizzi]

EDIT2: Whoever uploaded Smash T'ill You Crash 2 - it has the events but no matches entered, no pools, and a blank bracket full of bye's.

I know who uploaded it - I'll get in contact with him. Is the proper course of action to fix the issue and reupload it? Will you remove the incorrect entry?

 

[Varist]

EDIT: And Varist, please keep your trolling spam out of this topic. Everybody else is talking about how amazing this project is and how it will benefit our community.

Haven't posted yet, just been silently checking for updates since the thread started, but I have to do it now. This is ****ing awesome. Looking more and more legitimate.

When the day comes that these players are cataloged by region also, and you can split them into top-10s on that basis TOO, pretty sure collective underwear will need to be changed

sticky makes my pants sticky

ssbpd makes my socks sticky

but yeah if you just want everyone to keep praising your work Ziv I've got no problem saying it again. good job

edit: wait and HOLY **** i just realized you said this right after my post in the social thread. that was really sneaky, didn't even realize the parallel until after I posted this. CLEVER MAN.

and i don't act cool, I am coo right?

 

[Zivilyn Bane]

I know who uploaded it - I'll get in contact with him. Is the proper course of action to fix the issue and reupload it? Will you remove the incorrect entry?

Yeah, since it's not yet been processed his upload should overwrite the flawed file.

Varist: It's not my project so don't praise my work, praise FoxLisk. I'm just a clerical worker.

But yeah I was interpreting your post about Kels as a jab to the project and it being inaccurate. I didn't really look at your other posts. My bad.

 

[Rud Lisi]

I'm sure if mahone was here right now, he would be very proud of you alexander, and he would take full credit for getting this stickied even though i doubt they were even remotely related

 

[-ShadowPhoenix-]

Why is Cosmo not #3 ????

 

[ajp_anton]

OK, let me try to explain again how I'm thinking:

If Armada wins over Mango 3-2, currently the system only gives Armada a win and Mango a loss. This is OK. However, to get any kind of statistically valid data, they would need to play a lot more sets in order for us to see how often one wins over the other.
But if we take into account those 2 matches that Mango won, we get slightly more accurate results. From this 60/40 ratio for individual matches, we can calculate how often Armada should win if they played lots of best-of-x sets.
We can skip the individual matches if we want to, but they can give more accurate results *faster* than only counting sets, because we have a lot more matches than sets.

Now for the actual problem:
I'm assuming that the vast majority of the sets that have been uploaded to the system are best of 3. The ELO points should tell us the probability that a player wins over another *in a best of 3 set*.
If a best of 5 set is uploaded, that result is biased towards the better player, and will not give an accurate result. Sometimes the worse player should win, but this happens less frequently in bo5.
It's like counting how often you can score a 50 on Dart, but *sometimes* allowing multiple tries.

 

[Strong Badam]

Elo isn't supposed to show how often a player should win over another. Just a relative number that shows how consistent a player is. Elo doesn't factor in things like match-up disadvantages either.

 

[ajp_anton]

Um, that's exactly what it does.

If players A and B have ratings Ra and Rb, then the probability that player A wins over player B is:
p = 1 / (1 + 10^(-(Ra-Rb)/400))

 

[Kal]

ajp, the issue we take with counting wins, instead of only sets, is that in tournament your individual wins are unimportant. What matters is whether you win the set. So, it would be strange to have a tournament where "winning 3-2" and "winning 3-0" are identical, but to have rankings, based on said tournament, where they are not.

 

[Varist]

ajp, the issue we take with counting wins, instead of only sets, is that in tournament your individual wins are unimportant. What matters is whether you win the set. So, it would be strange to have a tournament where "winning 3-2" and "winning 3-0" are identical, but to have rankings, based on said tournament, where they are not.

I was agreeing with ajp until this post. Very logical explanation that sort of closes the book on the matter.

 

[stelzig]

Is it more strange than someone winning sets in pools/early in bracket 2-1 and then later on winning sets 3-2 after being behind 1-2 (or 0-2)

I just really don't see what you're getting at with that statement at all. I must agree with AJP as well.

Edit: It's supposed to be based on the tournament matches anyway, and not straight up tournament results. It would be just as weird to consider "your individual wins" as sets then. We could just look at placings and who they outplaced and who they didn't outplace.

 

[Fortress | Sveet]

A set is the atomic piece of a tournament, not a game. The TO doesnt go "oh u took a game off armada so you go to the loser bracket with 2 free wins". No. Just like it doesn't matter how many stocks you win with, a win is a win is a win. Period.