Introducing FightLines

by Mike Fagan on Jun 25, 2008 5:23 PM EDT Comment 25 comments

Very cool! Fronted by Kid Nate.

I started the FightLines project in late 2007. You can see its short lived life here. The idea stemmed from Beatpaths, which provides a ranking system for football teams.

After frustrating myself with an inability to successfully use the data from the FightLines graph into a reasonable power ranking, I lost interest in the project. However, this post from Kid Nate made something click in my head:

Honestly after poring over all these rankings - especially the ones like MMA-ELO and Fightmatrix that have "objective" scoring systems - I’m more convinced than ever that ranking fighters is an almost purely arbitrary exercise.

I realized that the data in the graph was useful in and of itself: as a graphical representation of where a fighter stands in his division.

So what exactly is FightLines and how does it work?

It's really very simple. The purpose is to provide a completely objective method of ranking. To do that, it's fed the most basic information - wins and losses. Using the results of the past 3 years, the data is put into hierarchical graphing software. The output showing a line from one fighter to another represents a win over that opponent.

The twist is when something like this happens:

Because you can't objectively say that any of the 3 fighters is above any other, the data points are removed or "looped" out. Loops are not limited to just 3 fighters as above. A loop can extend to 4, 5, 6, and more fighters.

If you check out the old FightLines blog above, you might notice that the graphs are unwieldy with large amounts of fighters involved. While I still am drawn to such massive looking images, for the sake of simplicity and practicality I've come up with a strict qualification system:

1) Any fighter on a legitimate top 10 list is included (with the recent comeback of the meta-rankings, I will be using them as my basis). If this provides an insufficient sample size, I use the Fightmatrix rankings to add additional fighters.

2) Any fighter who has fought 2+ fights against fighters in qualification 1 is included.

Here's the pilot graph for the light heavyweight division (too big for here, opens in new window).

And the loops are:

Arona = Silva
Liddell > Silva > Jardine > Liddell
Griffin > Bonnar > Jardine > Griffin

What does this graph show us?

To start, be careful about how you interpret the graph. For instance, Sam Hoger is not ranked higher than Forrest Griffin, even though he appears on a higher plane. This is simply how the graphing software lays out the nodes. The important thing is to see who is at the top/middle/bottom of each path.

For instance, Quinton Jackson, Lyoto Machida, Rashad Evans, and Thiago Silva are all at the top of their paths. Jackson and Machida's status are more clear cut, however, as their paths have 11 fighters underneath them while Evans and Thiago Silva have 8 and 3 respectively.

That being said, the FightLines system agrees with the notion that Lyoto Machida's next fight should be for the light heavyweight title. If Machida does not get a title shot next, the next best bout would appear to be Thiago Silva (though he is rumored to be fighting Wanderlei Silva) or Rashad Evans (who is fighting Chuck Liddell).

What are the potential after effects of the next two big light heavyweight bouts (Jackson/Griffin and Liddell/Evans)?

Jackson win: Does nothing for his standing as he already holds a path to Griffin (through Liddell via Ortiz).

Griffin win: Murks up the 205 division. Causes a Jackson > Liddell > Ortiz > Griffin > Jackson loop (Liddell only holds one win over Ortiz in the timeframe, so he loses his path over him). Chuck would essentially become a floating node. Ortiz would be stuck under Machida. Quinton would hold a lone path through Dan Henderson.

The Liddell/Evans fight has much broader implications. As you can see, they are both at or near the top of their respective paths. These paths are also pretty exlusive from one another. So, a win for either is huge (and a Liddell win also helps Quinton out as well) and would determine the next logical contender after Lyoto Machida.

There's more interesting data nuggets in there, but I don't want to overload the reader. This is definitely a work-in-progress and I would enjoy hearing comments, criticisms, and complaints. I'll also do my best to answer any questions (I'm sure my explanation of the idea and implementation wasn't exactly pristine).

The FanPosts are solely the subjective opinions of Bloody Elbow readers and do not necessarily reflect the views of Bloody Elbow editors or staff.

0 recs | Comment 25 comments

Story-email Email Printer Print

Comments

uh........

huh?

by strictlymma.com on Jun 25, 2008 5:39 PM EDT reply actions 0 recs

Is your goal to rank or to bet?
You are on to something, I just don’t know if that something is useful. I think you need a new method for dealing with loops. If your purpose is to rank fighters then you are fine in taking a method and developing rules based on wins and losses. Its an objective way to rank on a limited dataset (given that you need to address loops a little better). But if betting is your goal then you need to rethink it. A flow chart is not sufficent to capture the data needed to spot trends.

If you think about baseball and the batting title. They use average to find the best hitter. Does having the best average mean they are the best hitter, no. It is objective in that they decide the rules and the top guy wins. There is still a small facet of subjectivity in the hands of the scorer to say if a hit is infact a hit or an error, but over the length of a season the best hitters tend to bubble up to the top.

The problem with MMA is that guys don’t fight everyday and you can’t get a good average level of skill vs. a wide range of fighting styles. In baseball they solve the problem of limited data set with a minimum at bat number to be concidered for a batting title. In MMA you can’t make a guy fight everyday and you can’t assume he is the same fighter today that he was three years ago. You also can’t assume that a guy has a chance v. a BJJ guy because he out pointed 10 kick boxers by taking them to the ground and holding them there.

I don’t have an answer for your problems. I just understand how difficult the problem is. I remember your original work on this and its alway been in the back of my head.

by szucconi on Jun 25, 2008 5:40 PM EDT reply actions 0 recs

FightLines would be a terrible resource for gamblers :).

The sample size problem is huge for anyone trying to implement an objective system. That’s why a lot of ELO or similar systems don’t always look great.

The idea of FightLines isn’t to provide a specific 1-10 ranking (although the potential certainly is there). The biggest problem is that unlike Beatpaths, MMA fighters don’t fight the same small pool of talent that NFL teams do. So the graphs are a lot more ambiguous.

Still, I think it provides some useful information. I think it’s incredibly handy in helping determine how many fights away a fighter is from a title shot. It’s also a useful tool to help find interesting matchups.

I think the interest in this is kind of a niche thing, but I think those willing to try and understand it will see its utility.

by Mike Fagan on Jun 25, 2008 6:06 PM EDT up reply actions 0 recs

To comment on some other stuff…

re: styles and predictive qualities…

The FightLines system shouldn’t be used to try and predict a fight or performance. It is only used as a way to judge past results.

re: fighter isn’t the same as X years ago

True, but a football team in week 17 isn’t the same football team as in week 1. Fortunately, football has a convenient way to base your results (over the course of a season). So, I have to make an arbitrary decision on which results to include. I think 3 years is a pretty good standard to go by for a variety of reasons.

by Mike Fagan on Jun 25, 2008 6:16 PM EDT up reply actions 0 recs

I think it is interesting in analysis of past performance, but to what end? If it were to create a top 10, what would it be? If it were for match making then what matches would you make?

I just think its wrong to handle loops in this fasion. I do admit that it looks much cleaner and is makes sence to do it that way, I just think you are lossing to much data based on one match that may have been very close. Jardine > Liddell by split D, not that I don’t think Jardine won the fight it was just close. What would it look like if instead of dropping loops, you droped the oldest fight in the loop?

by szucconi on Jun 25, 2008 7:07 PM EDT up reply actions 0 recs

I think the utility

is to clarify rankings w/in promotions and also to suggest matchups.
Agree that more work needs to be done to account for loops.

by Kid Nate on Jun 25, 2008 9:00 PM EDT up reply actions 0 recs

If you don’t eliminate loops, you end up with arrows going UP instead of down, which not only doesn’t really provide any more information (since I list the loops after the graph), but would be very confusing.

by Mike Fagan on Jun 25, 2008 10:37 PM EDT up reply actions 0 recs

What about eliminating the oldest fight in each loop to break the loop? I was just curious about what that would look like. It might be pointless.

by szucconi on Jun 25, 2008 10:47 PM EDT up reply actions 0 recs

I went ahead and tried that out. Here are the results:

http://i5.photobucket.com/albums/y193/mikeypatriot/JuneLHWAlt.jpg

After taking out the Jardine/Bonnar fight, it created a 5 fighter loop with Irvin and Alexander which also had to be adjusted.

A problem with removing the last fight is that you will sometimes remove fights that are entirely relevant. I was worried that I couldn’t think of an example, but the Liddell > Silva > Jardine > Liddell loop is a perfect example. The oldest fight from that loop is less than 10 months ago.

by Mike Fagan on Jun 25, 2008 11:01 PM EDT up reply actions 0 recs

I see what you mean. Its very interesting to see it, but doesn’t really tell us anything useful. With the loops taken out you can see some potential match ups that may clear things up. If I were a match maker using this my goal would be to have fighters at equal levels fight and move up or fight to complete loops and clear out the chart. Like a rematch of Vitor Belfort and Tito Ortiz could do a lot to clean things up. Just the lapse of time getting that Ortiz split dec over Griffin will bubble him to the top and clear up a lot. Bonnar v. Jardine. That would be a good one.

by szucconi on Jun 25, 2008 11:24 PM EDT up reply actions 0 recs

Also, there’s a formula from Beatpaths that you can use to come up with a power ranking number. If you count each fighter below in a path as a “beatwin” and each fight above in a path as a “beatloss”, the formula looks like this:

(beatwins – beatlosses)/(beatwins+beatlosses+ambiguous loss) = Power ranking

Then you normalize the power ranking by adding one to it and then dividing the total by 2.

An ambiguous loss is a loss which has been looped out without another path to replace it. For instance, Keith Jardine has 2 loops, but only one ambiguous loss because Bonnar has a bath to him through Irvin via Alexander.

by Mike Fagan on Jun 25, 2008 10:49 PM EDT up reply actions 0 recs

The biggest problem being that at most a fighter might have 10-12 fights in 3 years, and most will have 3-5. Unbeaten fighter in that timeframe will always have a score of 1, which makes it hard to rank strictly using the formula.

When I first started up the blog, I tried to figure out a way to weight actual wins or fighters who fought more prolifically, and I could never come to a reasonable conclusion.

by Mike Fagan on Jun 25, 2008 11:04 PM EDT up reply actions 0 recs

So the long and short of the diagram is that the more paths a fighter has, the more accomplished a fighter he is. Is that right?

I don’t mean to dumb it down too far, but I’m not sure what the point is otherwise, since it seems to me the diagram of any division will require ample exposition for the circular relationships and the “Sam Hoger Effect” you mentioned. To me, that is counterintuitive to the goal of the diagram, but I suspect I am missing something.

by Brett Jones on Jun 25, 2008 6:15 PM EDT reply actions 0 recs

I wouldn’t necessarily say more “accomplished” so much as the data is more reliable. If Forrest beats Rampage and blows up that loop, it doesn’t mean that someone like Rashad is a better fighter than Jackson or Liddell, but we just don’t know how they stack up based on the results.

by Mike Fagan on Jun 25, 2008 6:26 PM EDT up reply actions 0 recs

One thing I forgot to note/mention.

Guys who are on a top 10 list may still not end up on a graph. For instance, Vlad Matyushenko is on some lists, but doesn’t have any fights with anyone else in the top 10. I suppose I could leave him as a floating node, but I just removed him from the list.

by Mike Fagan on Jun 25, 2008 6:32 PM EDT reply actions 0 recs

No

I would put him in and since one could claim that he hasn’t LOST to any of those other guys, he goes to the top!

Wait…

Don't believe a word I say, I don't train BJJ. -- TangleBones

by jemaleddin on Jun 26, 2008 12:12 PM EDT up reply actions 0 recs

I noticed that Wilson Gouveia and Gorjan Reljic are absent from the diagram. Their line goes directly above Lambert. Gouveia also lost to Jardine, though, so I’m not sure how you would handle that.

Luiz Cane is missing as well, and he’s another interesting one to place. Ahead of Lambert, but behind Irvin?

To once again address how I should read the diagram: I should understand that people at the top of a list featuring more fighters are ranked higher than people at the top of a list with only one or two fighters?

by Brett Jones on Jun 25, 2008 7:44 PM EDT reply actions 0 recs

Its mostly not about top or bottom. Its about beat paths and counts.

by szucconi on Jun 25, 2008 9:33 PM EDT up reply actions 0 recs

The notion that all fighter ranking systems are arbitrary is wrong. The criteria you use to rank them might be arbitrary, but as long as they’re applied objectively and identically to all fighters, the ranking itself isn’t arbitrary.

Some ranking systems have picked their criteria poorly and are thus don’t provide any useful information. I think both Fightmatrix and MMA-ELO do a good job in attempting to identify the subjective criteria that can be used to make an objective analysis and are generally more reliable when it comes to making predictions than other objective ranking sites (MMARanks) and the more subjective lists (Sherdog, MMAWeekly, MMAPlayground, etc.)

by George Lucas on Jun 25, 2008 9:10 PM EDT reply actions 0 recs

sure

some ranking systems are better than others, but when I used to overweight the sites with “objective” scoring criteria and I really looked at the enormous differences between MMA-ELO and FightMatrix’ rankings I concluded that ultimately they’re no more definitive than any other site and shouldn’t be over-weighted.
Personally I put a great deal of stock in Fightmatrix’ rankings.

by Kid Nate on Jun 25, 2008 11:20 PM EDT up reply actions 0 recs

I forgot to say that I also really like FightMatrix’s rankings.

by Mike Fagan on Jun 26, 2008 1:06 AM EDT up reply actions 0 recs

This may be a silly question, but does the 3 year time frame cause significant problems? I was thinking that, when a fight became more than 3 years old, its removal could really jumble up the graph. It’s a really interesting model. I can see where it could be useful when determining the next title contender.

by Cannon Jacques on Jun 25, 2008 11:12 PM EDT reply actions 0 recs

I wouldn’t say it causes significant “problems,” it’s just a part of the system. I could extend the time frame from 3 to 5 years, and you still have the same problem. You could try to include every single fight, but 1) you might be including fights from 10+ years ago which intuitively should have very little bearing on present rankings and 2) it creates a very large and unwieldy graph which is pretty impractical.

by Mike Fagan on Jun 26, 2008 1:11 AM EDT up reply actions 0 recs

I think 3 years is probably a pretty good time frame. Any longer a time frame would get crazy, and the older the fights, the less relevant they probably are. I was just sort of curious about the problems with the time variable. You’ve obviously put a lot of work into this model. Nice job.

by Cannon Jacques on Jun 26, 2008 8:03 AM EDT up reply actions 0 recs

But

Wait – the problem with the 3 years is that such a timeframe ought to include the people who were top 10 3 years ago as well, right? It’s a lot more data, but it would certainly provide more context, and therefore a more nuanced result.

It also might be impossible given the data available…

Don't believe a word I say, I don't train BJJ. -- TangleBones

by jemaleddin on Jun 26, 2008 12:14 PM EDT up reply actions 0 recs