What's new

Injustice: The True Top 10 Characters based off of tournament results by P2W (NA edition)

Why am in this? Last time I checked people that haven't made top 8 yet said I'm irrelevant in injustice.
You did at one point say that you didn't like this game and didn't put time into it like you did MK9. Are we supposed to believe you will still be a top player without passion or practice?
 

NoobHunter420

Scrub God Lord
Well if Chris G is the "God of Injustice" what happened to him last weekend at Summer Jam? Idk who but I heard there was some guy in top 8 that beat him like 5-0, if Chris G is a God than what is that guy? lol :D

lmao, then can we all agree that he is the most consistent player?
 

SaJa

FH_FenriR
The "true" top 10 indeed ^^

We can't say it's a true top ten just taking the winner final of something imo.

If I see 1 GA winning a whole tournament whereas there are tons of superman & BA players in the top 16, I won't say GA is as good as superman or BA. However I like those stats.

Would be cool to have such a thing for mk9. =)
 

JLG

Noob
Have there been over 100 entrants at a UK tourney? Not trying to instigate, generally curious.

I wonder what the character usage looks like for not only the UK, but also any place outside of the US. It would be interesting to see if certain cultures gravitate to different characters because of their aesthetic design, tactics, or whatever else.
 
Did someone say Objective top 5? (No, sadly no one did, but if they had here is the response).

This weekend, thanks in part to this thread, I took some time to calculate the ELO ranking for injustice based on the events in the OP (CEO, EVO, SJ, UFGT, ECT).

I tracked the win/loss ratio for 170 players across 223 matches and used a modified variation of the standard Chess ELO ratings to generate the following top 5 players using performance data. (I actually have the information together for 100+ players but more on that at a later date).

1. NYChrisG - 27 matches on record - 1541 ranking points.
2. REO - 20 matches on record - 1387 ranking points
3. Rico Suave - 12 matches on record - 1375 ranking points
4. Deg - 8 matches on record - 1356 ranking points
5. Perfect Legend - 15 matches on record - 1338 ranking points

Because of the nature of ELO rankings the more data I throw at it, the more accurate this list will be. However, as it stands the top 5 have an average confidence value of about 80% based on the current data set - which is not insignificant, but could be better.

Is it possible to estimate character strength using this type of approach? Perhaps, you would likely need to adjust the characters performance based on the player performance to attempt to isolate the strength of the character from the strength of the player - with enough data, and effort it might be possible. But today, with what I have it's not really feasible to create an objective, and mathematically sound model for character balance.
 

SaJa

FH_FenriR
Btw, did DD didn't win more major than anyone else ? I heard Pig saying that and I've seen PL win a major with him.
 
Did someone say Objective top 5? (No, sadly no one did, but if they had here is the response).

This weekend, thanks in part to this thread, I took some time to calculate the ELO ranking for injustice based on the events in the OP (CEO, EVO, SJ, UFGT, ECT).

I tracked the win/loss ratio for 170 players across 223 matches and used a modified variation of the standard Chess ELO ratings to generate the following top 5 players using performance data. (I actually have the information together for 100+ players but more on that at a later date).

1. NYChrisG - 27 matches on record - 1541 ranking points.
2. REO - 20 matches on record - 1387 ranking points
3. Rico Suave - 12 matches on record - 1375 ranking points
4. Deg - 8 matches on record - 1356 ranking points
5. Perfect Legend - 15 matches on record - 1338 ranking points

Because of the nature of ELO rankings the more data I throw at it, the more accurate this list will be. However, as it stands the top 5 have an average confidence value of about 80% based on the current data set - which is not insignificant, but could be better.

Is it possible to estimate character strength using this type of approach? Perhaps, you would likely need to adjust the characters performance based on the player performance to attempt to isolate the strength of the character from the strength of the player - with enough data, and effort it might be possible. But today, with what I have it's not really feasible to create an objective, and mathematically sound model for character balance.
this is interesting... i will send u a dm... i think u are counting individual matches when i think we should be counting the whole set
 
I am not one to ask things like this, but...

I won Evo, got 3rd at CEO and 9th at Summerjam.

How are most of those people above me elo ranking?
Something about the system may be a little off.
i agree... i think the elo system counts each individual match when in reality it should be counting entire sets
 
I am not one to ask things like this, but...

I won Evo, got 3rd at CEO and 9th at Summerjam.

How are most of those people above me elo ranking?
Something about the system may be a little off.

Firstly, I am still tuning the math so it's not out of the question that the system is off, although it's unlikely baring an outright error in a key part of the formula. This was just a test run, and was limited to those events in particular.

ELO is based on who you beat, who you lost too, how much you beat them by, and by how much the math predicted the winner to be. It doesn't care about what place you took. And by no means should it discredit your performance at any given event. Within the context of the event your place is the most important thing. But within the context of the larger ranking system who you won against and lost against is more important, and by how much.

Rough Example:
This is a match you played at CEO2013 against Deg
Going into the match the system predicted a 47% chance for you to win (Deg was on fire at that tournament which highlights one issue I have - I need more data). However, you took the match over Deg 2-1, essentially a win ratio of 66% outperforming the predicted results, this resulted in moderately decent boost in your rankings. Had the system predicted an 80% chance of you winning and you won by the same margin your rank would have move up VERY VERY slightly. Etc.

In short, KDZ (and again, mad respect) you lost matches that the math said you should win, you won matches that it said you should win, but you rarely won matches that it said you should loose. You took big hits when you lost to people ranked well bellow you when you fought them. Also, you had less matches in the reported match logs than some of the other fighters, which means some of your winning matches aren't reflected.

For this pass it was also very heavily limited by the reporting of matches, I really didn't have the time to watch all 223 over the weekend an instead used the results logs from Shoryuken. This meant that certain events had far more data than others. CEO2013 is an example of that.

There is another issue here in that I maintained objectivity by weighting all of the tournaments equally. For a variety of reasons I believe that a more accurate (but less objective) ELO system would have different weights for each tournament. Acknowledging things such as Patch changes and Meta changes.

More data will stabilize the ELO rankings.
 
i agree... i think the elo system counts each individual match when in reality it should be counting entire sets

It's actually counting the ratio of the games within the set.

The real issue is the lack of data. Take SJ7 for example, I only had the top 8. Which means all of KDZs matches at SJ7 are not factored in.

My plan is to surface all of this information in a searchable format on my site, but I need to do a bit of development first (searching the list for your name, finding matches and events, and submitting verifiable match data). I just wanted to put a small little preview up based on this threads content. My goal is complete transparency in the data.
 

KDZ

It's amore, BABY.
Are you sure?

At CEO, I lost to Deg in winners (who you said was statistically set to bet me, marginally), and then Perfect Legend. Both of whom are above me on this list right now. Then at Evo I went undefeated. Then at CEO I lost to khaotic (I guess the system would put that as an upset), and REO (also above me on the list).

According to your own logic, at some point I had to have been the favorite in these matches. But according to this system, I'm the underdog.
The Elo system might not favor this kind of play.

What I vote for is the ATP Tennis system, which nearly mirrors our tournament system (Since they use seeded brackets with individual play as well, albeit single elim). It is the closest, most accurate system for tournament-bracket-style INDIVIDUAL player ranking, and scaled based on player skill from previous results AND amount of players entered.

There's actually one for fighting game players that exists right now that uses that system.

We would be able to get it up and running almost immediately.
 
Are you sure?

At CEO, I lost to Deg in winners (who you said was statistically set to bet me, marginally), and then Perfect Legend. Both of whom are above me on this list right now. Then at Evo I went undefeated. Then at CEO I lost to khaotic (I guess the system would put that as an upset), and REO (also above me on the list).

According to your own logic, at some point I had to have been the favorite in these matches. But according to this system, I'm the underdog.
The Elo system might not favor this kind of play.

What I vote for is the ATP Tennis system, which nearly mirrors our tournament system (Since they use seeded brackets with individual play as well, albeit single elim). It is the closest, most accurate system for tournament-bracket-style INDIVIDUAL player ranking, and scaled based on player skill from previous results AND amount of players entered.

There's actually one for fighting game players that exists right now that uses that system.

We would be able to get it up and running almost immediately.

At CEO you are reported as fighting Deg multiple times. One was a bigger upset than the other.
VxG.EMP|KDZ (Superman) vs. Deg (Scorpion) – 2-1
VxG.EMP|KDZ (Superman) vs. Deg (Scorpion) – 0-2


For the other matches:
KDZ vs PL had you as the underdog, losing 2-3 (Down by 20 points)
KDZ vs REO had you as the underdog, winning 2-0 (Up by 30 points)
(I believe the Khaotik match was not reported)

As to ATP vs ELO that is a big topic, and a lot of people frown on ATP because it factors activity in so much. It also typically has a very high K value. This results in less stable results/data.

Generally stats view the systems as the following
ELO = Quality of results
ATP = Quantity of results

In other words, to move up in ATP you want LOTS of decent to good results, and regular placement is key. In ELO it cares more about who you beat, than how often you play.


Regardless of what system is used, and I'm happy to look at other systems, the data is going to be the killer.

I also don't like scaling based on number of entrants, the question isn't how many people played around you, but how did you do relative to the other players.

Some further data insight - because:
SJ7 only had 10 matches reported
EVO2013 had 70 matches reported
CEO had 64 matches reported
ECT5 had 55 matches reported
UFGT9 had 22 matches reported
 

KDZ

It's amore, BABY.
godlessmode

But now we're getting on the topic of someone getting a better score because he got a bunch of upsets and got 9th, vs someone who, due to those upsets, was the favorite to win, stays in winners and wins the major.

That seems suboptimal.
 
godlessmode

But now we're getting on the topic of someone getting a better score because he got a bunch of upsets and got 9th, vs someone who, due to those upsets, was the favorite to win, stays in winners and wins the major.

That seems suboptimal.

I get your point. But, this approach has proven very sound for every other sport and competitive system that uses it.

The issue is that the data is not fully available, and I am working on solutions for that.

As a hypothetical:
Who did better player A who beat no one of note on their way to take top 8, or player B who beat all of the top players on their way to top 8?

The problem with just basing the ranking on tournament results is that you ignore the level of competition faced within the tournament. This approach is about understanding the competition faced.
 

Saboteur-6

Filthy Casual
NOTE: This post is in response to the "definitive character tier list" portion of the OP.

Tl/Dr: Skewed data sampling is too inconclusive to properly claim a "definitive character tier list".

Not to get super nerdy but you can't correctly quanitfy the validity of your hypothesis (which is presumably that you can deduce a definitive tier list based solely on major tournament results) without adequate representation of all variables (in this case ALL of the characters of Injustice being played an EQUAL amount of time). It's for this reason that your data sample is biased and not objectively measured.

Basically...just because a character is played often in a tournament does not directly correlate to "character B is low-mid tier because they aren't used often/more often". All your data reinforces is the popularity and use of certain characters. That in and of itself is not adequate enough information to quantify something like a "tier list". Still though impressive work, especially with trying to calculate player rankings. Not trying to be a total downer.