I Revised my UFC Elo Engine...

[@GigachadMMA69]

belal p4p over jon lmao dana abt to end it all

[@keegswl5119]

Dana gad heart attack seeing that

[@dedde1311]

if belal was younger and could have more fights in the end he'd have fought better competition than jones ww is less shallow and more skilled than lhw then and lhw now

[@sticktothefacts8905]

@@dedde1311 If, if............. if is just making things up. Lets look at Jon Jones resume from the point he won the LHW title, Shogun Rua, Quinton Jackson, Lyoto Machida, Rashad Evans, Victor Belfort, Chael Sonnen, Alexander Gustaffson (twice), Glover Texeira, Daniel Cormier, Ovince ST Pierre, Anthony Smith, Thiago Santos, Dominick Reyes then moved to HW to beat Gane. There are 8 fighters there that either won the LHW or HW crown & 3 fighters in the Hall of Fame. The biggest nuff nuff in that list is Anthony Smith. Nobody has a resume that strong in the UFC, not even GSP or Islam Makhachev

[@keegswl5119]

@sticktothefacts8905 so a bunch of fat middleweights.... when Jones fought actual people his size he either scraped by or was caught on PEDs.... the only good win he has against someone his size is Gane and everyone knows Gane can't wrestle

[@Olabuenasnoiches]

​@@sticktothefacts8905saying gsp is less is crazy😂😂

[@mannymengs]

Still need that PED adjustment lol

[@nov3m472]

jon jones tested positive on a dickpill from the gas station a week prior and a party drug that had traces of peds because of bad labour. for both instances the dosis was so small it wouldntve been a positive test for todays usada standards.

[@riot5093]

it was a picogram :(

[@shako5761]

yk what nate said

[@mannymengs]

@@shako5761 true, adjust score on all the cheats. That’s why guys like GSP, DC, DJ are my goats (especially those first 2)

[@johnmarkson1990]

@@riot5093 that picogram saved his legacy. cause it took the attention of the other more extreme drug test failures.

[@ggthewhale]

Damn... Islam might become goat of the sport

[@hissy7042]

If he defends against Arman and becomes Double champ he’s for sure second behind jones. Of course he doesn’t have the quantity of title defences gsp had but the quality of opponents he has faced has been far far better

[@olisk-jy9rz]

@@hissy7042 Nope. GSP's resume is still miles better. He also became champ in two weight classes, he didn't rely on extreme dehydration and weight/strenght advantage the dagestanis fighters do. Islam needs to become champion at welterweight to be considered in the GOAT discussion.

[@dedde1311]

@@hissy7042 The competition islam faced is already better than any jones win other than dc that single dc win is carrying his legacy, as it's his only elite win

[@Harinjo916]

​@@dedde1311Fun fact Jones tested positive 3 times against DC alone over whom he has reach advantage, height advantage (18cm) and DC was already washed up, he was 37 If I remember correctly, Jon only fought 3 guys younger then himself, Reyes was one of em. The rest were @ 36 y old or around it

[@Figgy636]

@@olisk-jy9rz GSP was a huge welterweight and clearly much physically stronger than most of the division. I don’t see the difference. Anyway Chandler is the same size as Islam. There are plenty of guys who actually miss weight when Dagestanis don’t. So your criticism doesn’t mean much

[@Dwild2964]

Mighty Mouse at 15 all time peak elo is interesting.. also Beneil Dariush above Amanda Nunes haha

[@WhitePillMan]

He faced weak competition. Nothing he could do about that.

[@PAWGSOTHOTH]

The tiny dudes don't move weight classes and are overall the weakest fighters. They are skilled, but they still get smashed by a dude 20 pounds heavier who is similarly elite. A guy with a larger frame can jump around and has better opportunity. Dj's essentially king of the bottom feeders, and I'd wager 60% of elo sits between 145-185 (half the divisions)

[@dopaz9616]

The Flyweight division is small, so he has fewer opponents as well as lower elo opponents.

[@OnPaperMMA]

@@WhitePillManthat’s quite a misnomer in my opinion. He faced some incredibly good guys, just made it look easy most of the time. Dodson, Benavidez, Cejudo, Horiguchi are all elite level wins, just not as well known. I think what really affects his elo is the fact that he rarely ever won a fight where he wasn’t a large favorite, because of how dominant his reign was, and how quickly he rose to the top due to the fact that has was part of the original 2 fight flyweight tournament

[@WhitePillMan]

@@OnPaperMMA I would like to agree w you but favorite shouldn’t affect his ELO. His ELO would be based on who his opponents beat in their careers, not the gap between Mighty and them.

[@Philbobama]

My guy. You saw my comment, hearted it, and included the reply as a section of this video. That’s so cool of you, thank you bro, that so awesome. Great content

[@PulsatexFf]

Send those to jre hes gonna love it and shout you out

[@darkability1393]

Something interesting is I tried including other promotions (Bellator, Strikeforce, PRIDE, WEC) (so things like Jon Jones' ELO will be bumped higher since he fought a lot of PRIDE guys), the current top 5 is now: Jon Jones,1316.02 Islam Makhachev,1273.52 AJ McKee Jr.,1273.26 Patricio Pitbull,1271.43 Ryan Bader,1261.03 I don't watch Bellator so no clue who AJ McKee Jr or Patricio Pitbull are, I found it interesting that 3 of the top 5 are from Bellator though. If we do all time, the top 10 is: Jon Jones,1316.02 Daniel Cormier,1287.89 Patricio Pitbull,1287.26 Islam Makhachev,1273.52 AJ McKee Jr.,1273.26 Gegard Mousasi,1271.77 A. Nogueira,1265.75 Donald Cerrone,1262.07 Fedor Emelianenko,1261.06 Ryan Bader,1261.03 With Anderson Silva not even making the top 50, as he barely misses it, I'm assuming due to the people he fought having less time to build up a high ELO? If I bump up the K value to the one shown in the video of 120, these are the top people: Islam Makhachev,1751.63 Jon Jones,1744.32 Leon Edwards,1712.53 Kamaru Usman,1709.85 Daniel Cormier,1689.25 Belal Muhammad,1687.66 A. Volkanovski,1683.12 Charles Oliveira,1680.55 K. Nurmagomedov,1677.89 Stipe Miocic,1669.0 Aljamain Sterling,1660.44 Patricio Pitbull,1647.15 With Islam Makhachev beating Jones. Very fun

[@TrixsterProductions]

That is very interesting!

[@kije6639]

You should strongly consider lowering the K value substantially for Bellator specifically

[@user-go2xi7zq5q]

This is a solid series bro. Keep on. Maybe set it to automatically update, based on new results in the database. And put it on a website somewhere. Maybe email Verdict MMA app, and tell them to add it somewhere on the app section, giving credit to you. Goodluck bro👏🏾🐺

[@XTSonic]

At 3:24, why is GSP greyed out. He ain't dead lol

[@TrixsterProductions]

I couldn’t find a better png, and im lazy to mask out regular images bro😭

[@shivamsrivastava6312]

LOL

[@polskillz]

bro is a locked character

[@BacatauMania]

More important than boosting submissions and knockouts would be making the points gained/lost relative to the judge scoring difference. A 50-45 should give more points than a 48-47. But there's something I didn't quite get which is also very important, beating Jon Jones gives more points than beating fucking Ben askren, right?

[@TrixsterProductions]

I see the point for the first one, not sure if I’ll be doing a third update on this particular thing but I’ll update the code and results on the git fs. And well yea I guess it should, but it also depends on the matchup for each fighter at the same time

[@BacatauMania]

@@TrixsterProductions I think the most balanced way of doing things would be starting every single fighter with the same rating, but simulating the fiights chronologically. That way, new fighters can farm more points beating older guys which have higher elo points, like a 1000 adesanya beating a 1200 Silva. New generation being propelled based on their opponents history. It sounds like a universal solution to me, maybe not as easy to execute.

[@meateaterscringe9863]

@@BacatauMania I think this is exactly how elo works.

[@kono5933]

​@@meateaterscringe9863It is

[@BacatauMania]

@@meateaterscringe9863 I thought too, but when the author of the script said beating Jones and askren would give the same elo points, it made clear how it's not the case

[@kidMMA4000]

Imagine how amazing it would be if you could make an engine that is almost undeniably perfect that takes every important factor into consideration,

[@raphaelfrach5397]

But that isn't really what Elo is about, in chess it is just wins, losses and draws. But I guess chess is a much more standardised game

[@brusombear3823]

@@raphaelfrach5397 i think finishes should count for more in mma

[@Ebaadsiddique]

I don't know if I commented this in the last video, but I'll do it here. UFC is almost definitely the highest level of competition of mma, but wins and losses in other organizations could change results slightly. For example, Micheal Chandler and Michael "Venom" Page were successful in Bellator. On the other hand, Anderson Silva lost some of his early mma matches. Fedor Emelianenko is considered one of the greatest heavyweights of all time but never stepped in the UFC. Demetrius Johnson's career continued in Prime all the way until his retirement just this year. I know UFC stands out when it comes to data collection, so it may be hard to implement this into the ELO system, but it would really impressive to see.

[@TrixsterProductions]

And that’s something I’d be interested in, but it may be a bit later on , I still need to scrape a lot of data from like sherdog/ figure which fighters to get data for, maybe I scrape data from all bellator and One championship, and then merge everything together some how, but that requires a lot of time to figure out and execute haha. But I am definitely interested in doing that

[@sticktothefacts8905]

Anderson Silva didnt lose a lot of his early MMA matches. He was 17-4 when he got to the UFC and one of those was a DQ for an illegal upkick. 2 were in Pride & the other was his third ever fight.

[@Ebaadsiddique]

@@sticktothefacts8905 I appreciate the correction

[@WindsorHornIII]

@@TrixsterProductions forget bellator, pride and strikeforce had much more important matches for the mma history books

[@TrixsterProductions]

@@WindsorHornIII true true

[@MrBleedforme]

Tony is the type of person who's going to peek at a peak list.

[@emssasukeisunderrated7946]

This is quickly becoming my favorite type of videos! As someone into analytics and c.s. this warms my heart haha. But it also gets me thinking of ways to make the model better. I would also think the k-value should change but finding the right value as for it to not be very volatile. It seemed like when the k-value raised up to 200 the output had a similar issue to the original which was giving more advantage to current fighters. Could be cause the ones who are champs are peaking at this exact moment but who knows. But below I'll list some factors to add in. The great thing about a dataset, which this data looks to be a decent size, is there's ways to test the model and make predictions regarding the data (i.e. when do fighters peak, what is the mean elo of first time champs, etc.). Anyways here's more factors to include to better fine tune the elo model: 1. Rounds: not every fight is the same round. Winning a 5 round fight should be weighed more than winning a 3 round fight especially since 5 round fights typically are only for co-main, main, BMF or real championship fights which includes my second (and most important factor) 2. Championship fights: Winning a championship fight HAS to have more weight than a normal fight. Also losing a championship fight should NOT be weighed as a detriment as much. Also, if both the champ and challenger are both former interim/unified champ (AT THE TIME OF THE FIGHT) should matter too and should apply the prior sentence even more. Take Izzy and Poatan. Izzy beating Poatan when both have already been champs should be a higher factor for Izzy having 1) KO'd Alex 2) winning in a title fight and 3) beating a current/former champ. For Alex the only significant detriment is him getting KO'd 3. Decision fights: Granted, I'll admit this one might not be necessary but winning by split decision is much different than a 30-27 or 50-45 dominant massacre. This is of the same ilk as separating KO/Sub and winning by decision I'm very interested to see how this will change the elo ratings. My hypothesis is the peak elo of the top guys will be much higher and this will favor guys who dominated their division for a long time. I would predict that Jon Jones, GSP, Khabib, Kamaru and Silva or Izzy (maybe even Volk) would have the highest peak Elo given how they all nearly (or did) clean out divisions and with how many times they fought in championship fights and against former/current champs at the time of the fights.

[@emssasukeisunderrated7946]

On another note, I'm interested in maybe seeing how another system such as Glicko/Glicko-2 as well. I know in a team sport or sport with offense & defense it would be better to use a more in-depth system. Example is using SRS or DVOA and weight-adjusting the values for eras (which maybe you can weight adjust for UFC although it may not be necessary since title fights and fights against former/current champs would the weight adjustments)

[@nicolasjoelhernandezduran1333]

I also think split decisions should give less points than winning with a dominant 50-45. The extra points for winning over a former/current champ is interesting, but maybe it should only count if you win in a top contender or championship fight. With this adjustments I also think dj would be in the peak ELO rankings if you take into account that he had like 11 defenses

[@joseandres32_]

If you post any content related to this, I'd definetely check it out. Those are great suggestions!

[@callidus9421]

It's interesting to see Khabib so low despite having won all his fights, as his Elo-performance has to be better that of all not undefeated fighters. I guess his opponents' rating average must be pretty low from before his title run.

[@CaptainBeano-lz6mm]

His whole career was a title run and 95% of the wins were against negative record journeymen. He only fought around 5 real contenders

[@mediumchicago_pizza]

@@CaptainBeano-lz6mmmore like 6 but i get where u coming from

[@CaptainBeano-lz6mm]

@@mediumchicago_pizza i dont count al iaquinta or mike johnson as real contenders

[@mediumchicago_pizza]

@@CaptainBeano-lz6mm at the time mj was def a top contender lol, dude had a freaky prime then couldnt adjust his game when his physical gifts started failing him

[@AstroKidLo]

@@mediumchicago_pizzahe said “around 5”… I think 6 qualifies for being around 5

[@genericname2821]

I actually really like this because it shows a new perspective about losses, a lot of MMA fans just discard a lot of earlier and later losses in a fighters career, even when a fighter is in his prime like in Jose Aldos case

[@SugaMan96]

Using this you can hava a ranking for all fighters in a division. And compere that to the ufc ranking. Thats a video i will watch

[@tabernadetebes5547]

you should multiply by 0 their elo if thei failed any drug test.

[@heroic5680]

suck ur own ass , loser

[@ericstewart-ce7eq]

Unfortunately that isn’t part of the data set lol

[@bobjames1521]

Every fighter is on drugs. Get over it.

[@atomandroid7232]

@@bobjames1521 You can tell when they are, acne, gyno, failed tests, sudden ko power, sudden bodybuilder physique. Volk, gsp and demetrius are pretty clean.

[@FeatherInABook]

​​@@atomandroid7232Depends how much and what they're doping with, almost all A-level sports athletes are using test or growth hormone, rarely with side effects under close supervision.

[@DasGuntLord01]

Rather than adjusting the K-value, you can actually adjust the *score* that the fighter receives! After all, if 1 is a complete win and 0 is a complete loss, then you can call a decision win as almost a draw (like 0.5 < score < 0.6, which I made up) and/or based on finish time, like winning at 0 seconds of the first round is a 1 by definition, and that slides down to 0.5 for a draw at 15:00? You make the scores up to your taste, of course, but the gist is that you can assume that quick wins are more dominant (on average), and then this system rewards dominance.

[@Arcdelio]

Really cool concept

[@danamesmumei]

you should also take avenged losses into account, and add a bonus for how said fighter avenged the loss.

[@bot6349]

BANGER!!! Subbed! Great follow up to the original video. A great idea I saw is maybe trying to incorporate how good someone ends up being into the ratings / winnings. it's a bit hard but my idea is to give extra winnings to fighters if they beat a future top 10, 5, or champ. Maybe try to make or find a function to find the optimal K-value for UFC purposes? There has to be code for something like that somewhere... What if fighters that have gone on losing streaks or will go on losing streaks give less ELO. Maybe reduce points gained to the winner just before (or not, maybe he caused the fall off and deserves it to be reflected) and all during the losing that benefit(so reduce points gained when Izzy beat Silva or anyone that cooked Tony after Gaethje). Maybe Jon will get outpoint by GSP in this version? Or Kamaru might drop out the top 5? Another idea is to reduce points for fighting someone from a lower division or smaller than you and more points for fighting someone bigger or from a higher division... This also has a benefit for someone the ELO engine has been overlooking, Mighty Mouse. As for Islam, he simply is better than Khabib. He has more notable wins against tougher, more versatile fighters, and he's also more complete as a MMA fighter with his crispy strikes. He arguably also fought some of Khabib's best wins when they got better, Poirier was a monster versus Islam and I think a lot of fighters are tougher now because we now know how good a wrestling base is and how important good takedown defence is. Of course this all because of Khabib, he is the pioneer. Islam wouldn't be where he is without Khabib and his father, but career-wise, Islam is better.

[@yellonuim1741]

great video, as a chess player, its kinda amazing how people's general opinions of best fighters align with elo ratings.

[@1337strvids]

Didn't think you'd go through the trouble of actually implementing the changes, definitely subbing great work.

[@FrayMMA]

Things that would be cool to see: Highest dropdown Highest dropdown before peaking higher!? Quickest rise/fall Love the data brother, keep those video's rolling.

[@dougdupont6134]

Man I made an elo for the UFC years ago (without even knowing what an elo was) in C# including an interface and everything. BUT the web scraping escaped me. I tried to learn R for that, but I just couldn't combine the two and gave up after months of effort. I was about to come here and beg for the dataset when I saw that you provided it already. You are the man. I'm going to make this into an automated betting system. If I make any money, I'll be looking for your patreon to send you a handsome finder's fee.

[@sam.whiting]

Sick video! I'd like to see it adjusted to take into account performance over time, so more activity is rewarded and a longer career is rewarded with consistency over time being important. It would also be interesting to see if you could work out the strike differential and control time differential as I think that would take into account who is more dominant during their fights... I think that's why Khabib is put so high on people's GOAT lists: he pretty much never lost a round if I remember correctly! Love these videos super cool idea keep it up!

[@davidlariviere8819]

this is great. plz keep refining it, we will all keep watching

[@richielavey1565]

I did something similar to this a while back and your videos inspired me to take a look at it again. Some things I learned that increase the auc (a statistical metric that shows the predictive power of the model) were: introducing a dynamic k-factor so fighters that haven’t fought in the ufc before have a more volatile score to more quickly get to their “correct” score, adding a decay feature so fighters lose some elo as time goes on and they get older, and comparing your results to the betting odds can give you a good idea of how good your model is compared to the pros! I was also thinking about doing some team sports next, reach out if you’d like to collaborate on that!

[@paladico]

Amazing! I'll leave my suggestions from the previous video if you want to work on a v3 in this case lol: - Add some depreciation mechanic to adjust for inactivity. - Maybe use age or time in the ufc (or both) as weights for K to adjust for where in their career a fighter is. - Maybe you can also use the amount of fighters in the weightclass at the time of the bout as weights as well for K. To penalize less competitive divisions. - You could also have a sort of buffer calculation for a fighters first fight. Or have two parallel elos one that starts from 1000 and another that starts from the elo of the debutant's first rival. And use an average of both until they converge. Loved the content if any of this sounds interesting and you would like to chat about it, hit me up.

[@zahvage1204]

You should keep making parts and keep them more and more intricate! Like next part account for how many rounds won Then next part account for take downs then next part for how many hits landed then .. i can't think of anything more

[@SharkiieOW]

Love these elo videos!

[@musicplaylists59]

an interesting way to change it would be to make it so they get more points for beating people when they are at their peak, so you gain less points for a win the further your opponent was from their all time peak elo.

[@FelixakaJopokakafa]

Not really a UFC fan, but huge data + nba + f1 fan, so mixing in those would be great.

[@rasifude]

I would just make one change. Championship fights should count more. But beautiful work!

[@Strikerborn]

Add different K factors for different types of finishes/methods of victories. Use Optuna to hypertune the K factors to highest accuracy using a backtest method.

[@alexeypereverzev3392]

there are a lot of other things that affects: fights at short notice, injures, moving to other weight classes, winning on the opponent's territory etc. So it always will be a fantasy game

[@OrignalElidest]

This is such a cool concept, so many lil things you can changer personally i think setting K to 50 would be a nice lil change and having something for early round finishes should grant more. Tom Aspinal hasnt been very high in the lists but the guy has spent less than 41 minutes in the cage in his ENITRE mma career and has the lowest fight average in the ufc. I think having that quicker fight finishes affecting elo could be pretty huge especially since jones went on a bunch of decisions in the later part of his career.

[@pietroghsvf]

I really like k-ajusted peak elo list, I think it includes all the legacy fighters in their correct range, I never thought Khabib was top5, top10 is good considering he retired early and could have fought some tougher opponents. Same for mighty mouse, he was dominant in the least competitive division with the fewer fighters, top 15 is good for him.

[@r6isbest836]

An elo rating for all the sport idea is amazing thanks for all the hard work keep at it!

[@strigiformsW]

Hey I really like this idea. You've earned my sub. I've wanted to do some data projects with UFC for so long but my masters research has become all consuming. I'd love to see more stuff like this.

[@ugopaleni2131]

Yeaaaah. That’s cool you went ahead with it so quick. Thanks.

[@Borgathan]

as someone who loves data and coding and also casually watches UFC I'm loving the videos. would LOVE so much to see some NBA content

[@TrixsterProductions]

😁

[@_DioBrando_]

If we're adjusting for finishes then we should also adjust for dominant decisions, 50-45 UDs, 10-8 rounds etc

[@jahcode6132]

I liked this video and I'm a huge NBA fan so I'm looking forward to see what basketball content you can make. Got my sub for sure.

[@RJavier007]

Is criminal that we did not saw Tony vs Khabib at there prime.

[@danielmaske9946]

I agree that peak Elo is better for all-time rankings than current Elo, so I'm glad you reformatted. Going by current Elo in chess I am actually higher than Fischer, Tal, or Capablanca.

[@lukeloobey]

First and foremost this is incredible and I love looking at it like this, as a comparison of 'resumes' that we often see talked about but usually with relatively little data involved. I'm also interested to see a list of values with different K values attributed per match, which if I understand ELO correctly means that winning is more rewarded and losing is more punished with a higher K value where a lower K value means consistently winning is more rewarded than an individual win over an individual high-score aka big-name fighter. Seeing a K-score matching chess's standard format, 20, 40, 100, and 200 would be very interesting to see what it means overall. Also I wonder how their scores would look if you did a hot-take controversy version, swapping Jone's win over Reyes to a loss and swapping GSP's win over Hendricks to a loss, for example, where the majority of polls and media outlets had the other guy winning as the defining feature of what qualifies as this special variant or something. Doubtful that single blip even reversed would knock those two off of the top of the podium but again, curious to see the alternate-future data numbers. Double-also, MMA On Point often does huge dataset breakdowns like this with Tommy Toehold and a collab would be dope.

[@TokyoAlex]

I love the idea of applying ELO to MMA. Some things that I would like to see added to the equation is some kind of depreciation for inactivity and another depreciation score for age. Someone like Stipe who is both inactive and old should be losing ELO everyday that passes. If you get the formula down, I would like to see it applied to real rankings and matchmaking... if that is something you are interested in. If not, I still appreciate your work. Very fun.

[@TheI3utcher]

I just found part 2! What a great turn around time!

[@shanehummusunfiltered]

So cool. One suggestion would be giving fighters a boost if they go up in weight and win (and also less penalty for going up in weight and losing)

[@ttipun]

If you can find the data, it would be cool to know the martial arts background of each fighter and look at which practice (BJJ, wrestling, kickboxing) dominates at different points in time-how the meta changes if you will.

[@amirnili6460]

another suggestion that might be a little technically challenging: pull the decisions of each round, and treat each round as a separate fight. That way you will get more data per fighter and you will be able to distinguish between a 30-27 and a 30-24. Or a 29-18 heading to the third round, but then the fight ending with a finish

[@zarifinator1327]

Theres not enough data on that. Theres date on scores for fights but round by round scores aren't really fully available. Especially for earlier fights.

[@TheFuriator]

This makes complete sense, it's a great list imo. Also Charles is goated

[@zapid6733]

Should try to add in results from different promotions as well, should be easy when it’s MMA like Pride or PFL but could also do it with other sports like kickboxing so if they have a good record there with big orgs like K1 or Glory you could have them start with a higher elo or K value dependant by their elo they had in the other sport and how successful other people transitioning from that sport have been. Maybe dependant on avg. betting odds on the fight. Could maybe add that as a slightly weighted factor as well if you were a huge underdog irl despite having same elo in your engine they could be weighted so both gain/lose more so it equalises.

[@404-Firefly-NotFound]

3:46 and separate scores based on submission or knockout so that we know whos the highest rated in grappling and in striking separately 😎

[@MrNegr43]

I did not see the dataset itself, but from the first video it seems that non-UFC fights do not count, which sort of penalizes athletes who had less fights in the UFC before (or after) joining it, like DJ or Askren. Also I would suggest assigning initial ELO as per bookies coefficients against first ufc opponent rather than making it equal for all, e.g. a newcomer having bookies favours in a first ufc fight gets opponents ELO + adjusted for coefficient bonus (let’s say newcomer has -300 odds as per bookies and this gives them initial ELO equal to opponent’s ELO plus 300*k_favourite). Anyways awesome job and looks really interesting!

[@alekseiaminev4751]

Awesome analysis!

[@TRex-fu7bt]

Increase K when fighters are crossing from their regular weight class 🤔??? Just thinking aloud. Still playing with my R fork of the code

[@dougdupont6134]

I'd probably consider weight class changes as a separate fighter ... it's too big of a variable. K value should probably be higher in new fighters, since they all begin with a base value that is only reflective of their UFC experience, not winning ability. It should also increase with inactivity.

[@drunkenswordsman3373]

Not a huge sports fan, but this is pretty cool! Keep up the good work

[@supergowiak4369]

You should do elo rankings of all numbered fighters per division. I would definitly watch that

[@thomasbyrnes3043]

I think it would be interesting if a UD shutout (30-27, 50-45) gave a finish multiplier, while a mixed shutout scorecard (48-47, 50,-45, 50-45) gave less of a multiplier, razor close decisions give no multiplier, and split decisions gave significantly less elo, as a lot of SDs are considered robberies (maybe .75) Im sure an algorithm can be concocted to factor how close a decision is on paper, so JJ and Mak don't get crazy elo from Reyes or Volk 1 but gain even more from fights like Gane and Volk 2 Edit: if you also modify the k-factor when one fighter is moving up/down the weightclass that would probably be a crazy good engine too

[@davedavy5406]

I'd be interested to see Pride vs UFC Elos from back in the day when they were both running. See how Fedor peak ELO compares with Silva's for example

[@Awesummzzz]

Mr V's Garage made an F1 elo video a while back. Recommend his content in general too

[@victorsilvapacheco9630]

Yesss!! Great video!!! Thanks!!!!

[@marekheczko9115]

Good job bro. I think split decision should be worthy of less point than majority descion and majority decision should be worthy less point than unanimous decision.

[@Carlos_cheh]

you definitely need to add something about the peds, definitely do

[@RyanMMA]

I'd love to see more combat sports, and NBA content!

[@Briney_Is_Cool_I_think]

My boy Trixter moving up in the world! Love these new ufc vids

[@TrixsterProductions]

😊

[@TrixsterProductions]

Thank u briney 🫶🫶

[@BandoLando]

thanks for the quick turn around on the video

[@FA-ul4pr]

I am liking this series. Please adjust the ratings so we can know who the super grandmasters are, the grandmasters and the regular masters and the international masters. Also, make your adjustment consider the length of the fight. E.g Alex KOs Hill in round 2 isn't the same as Alex KOs Roundtree in round 4 after a tough fight.

[@katlynklassen809]

I think an elo tyoe system with rules around regular title defences would create a truly fair and balanced system for fighter match making.

[@christianheath718]

I’m not sure how you’d apply it but I think having some king of adjustment based on the difficulty of the competitor should be added. For example, if someone beat 10 solid fighters and lost to 1 amazing fighter, I’m pretty sure they’d be lower than someone who just beat 10 mediocre fighters but never lost. Like I don’t feel like Mighty Mouse losing to Henry Cejudo or Alex Pereira losing to Adesanya should impact them as much since it’s just such a difficult fight. Or for example let’s say someone with a really high ELO loses to someone else with a really high but slightly less ELO, I think a loss to a great opponent should be mitigated so that it’s not much worse than a win over a mediocre opponent.

[@angelopueyygarcia43]

Islam is in a GOAT worthy run so he def better than Khabib. A couple more defenses and maybe a second title, and Islam would be in the GOAT discussion not just light weight GOAT

[@fightme5543]

I'd love to see an actual plotted graph displaying the journey that everyones ELO's took! You'll probably be able to spot key moments in fighters careers, major wins and losses, slow climbs and 'dana privilege' climbs!

[@graysonalt5225]

You should try and weight it by divisions. For instance heavyweights have way higher chance of getting a ko so its not as impressive as a ko at flyweight. If possible you should look at rates of submission by weight class and use that weight to compare which goat performances are more impressive statistically speaking. This is amazing that you've made this I have been wanting and thinking about doing this for a while. This is awesome keep it up man.

[@TrixsterProductions]

I see that yea, when I did the k value boost nganou like sprang up a bit

[@skegior2323]

You should use Sherdog, its got every single fight a fighter has and start from there, its hard to do alot of fighters so just make a scraper that scrapes data off sherdog and uses that, you could do individual fighters and make it so it calculates the elo of the opponents at the exact time of the fight so it gives you a 100% accurate elo score, which could also expand your data into other promotions etc

[@joshuabela5374]

Also, you can use machine learning to optimize the K-value. The optimal K-value should minimize the the average ELO volatility for all fighters, across time. The idea is, an accurate K-value should more reliably predict fight results, resulting in smaller adjustments to the ELO rating over time.

[@lambslide_]

as a big Robert Whittaker fan, i’d love to see what he scores if he starts at 1000 again after moving to middleweight tbh, I feel like different weight class should kinda wipe everything, since the point of going up in weight is a fresh start anyway loved the two videos I’ve seen so far though, thanks man ❤

[@signastytv5954]

Awesome video and follow up

[@Javix0000]

Great Job! I wonder if the system could be made to penalize inactivity in some way to impact fighters that are not taking at least one fight every 6 months or so.

[@randomthingsxd9030]

Definitely do basketball Settle the goast debate. Although team sport individual player elo is a lot harder

[@TrixsterProductions]

Definitely

[@9malklug939]

Adjusting towards Titel win and defenses aswell as being unbeaten are significant factors in terms of GOAT status. Might aswell include it somehow

[@kidMMA4000]

Part 3!!!!!!! Maybe make it so the code accounts for how good the fighters someone beat turned out to be, rather than just the current betting odds.

[@hartfordwolfpack8678]

That's like how UTR works I think

[@andrewnormand7787]

You could leave the results on the screen for more time so we can look at it😂 great vid tho!

[@David-bn2yr]

i think a way to improve this sistem a lot is to implement data of before ufc so the ones that start as champions contenders in the firsts UFCs, who are probably at the peak of their carrer, dont count as debutants with 1000 points. this should make the sistem more acurate because whoever beated the first ufc champion of their division did not beat a 1000elo but a 1250 elo or whatever. Also this way we could compare the gratest of diferent eras like Fedor Emelianenko and jon jones.

[@YappingForChrist]

Let’s go another one!!

[@infrieser]

Have you considered optimizing the value of K by approaching this like a machine learning algorithm? You can use a player's current elo at time t to predict the outcome of an upcoming match at time t+1. This way you can view the value K like a hyperparameter that can be optimized with cross-validation, yielding more accurate elo scores.

[@jakebishop7822]

That is indeed a cool update. If I keep giving suggestions will they keep happening? What if instead of going fight by fight we went round by round to calculate elo. with a 10-8 getting a slight boost to how dominant of a round it was, and a finish getting an even bigger boost This might not be as doable because as far as I know the information is not as well organized, so it may not be doable, but right now the biggest flaw in elo for evaluating fighters is that a close win is the same as a dominant win.

[@kono5933]

And is not being able to find the 'true' value of k, or also how much a finish/more dominant round should actually increase the Elo gain/loss by In theory if you used machine learning you could probably test which values more accurately predict the outcomes of future fights, and then we'd have more accurate rankings

[@giorgiochiodini5773]

let me understand: if i win by ko on round 1 i get less points than win 4 rounds and close by ko on round 5?

[@jakebishop7822]

@@giorgiochiodini5773 i didn't specify the boost of getting a finish, but what you could do is make a finish the equivalent of winning every remaining round of the fight 10-8. Which would make a first round finish the best possible result.

[@SFTMproductions]

ITS TONY TIME ‼️‼️🗣️

[@fahed2544]

Great video

[@elijah1163]

Suggestion: do "performance rating" instead of elo. In it's simplest terms, performance rating is an average of each bout, where a win gives you a performance of [opponent's rating + 400], a draw gives you a performance of [opponent's rating], and a loss [opponent's rating - 400]. I would suggest ranking finishes with the +-400 value, and maybe unanimous decisions +-250 and splitty's +-100? Performance rating would better judge a fighter's ability without rewarding their total activity. I think this method would be great for adding to the discourse on No1 contendership. Pleeeease do it ahahaha love your content thx.

[@elijah1163]

Would be awesome as well to have a data set that show's judge's scoring. Then you could differentiate performance based on the point spread. For example, if each point spread gave you +-27, then 50-45s across the board would net you +-405, 30-27s would net you +-243, 29-28s would give you +-81 and a 29-28 splitty would give you just +-27 etc. This would do a nice job imo of tracking dominance and estimating overall fighting prowess.

[@mylespritchett7957]

One thing I am interested in is adjusting the elo gained depending on the fighters future fights, not just their previous fights. For example, Charles Oliveria beating Tony Ferguson would still have won him a significant amount of ELO, but looking back we can see that Tony lost all of his fights after this, so maybe that win was not as important is it should have been. Not sure what the best way to influence this would be, maybe for their next 4 fights, the outcome influences your elo by 20, 15, 10, 5% either gains or losses. If you lose to a lower rated up and comer that then dominates everyone else in the division it is obvious that win should not have taken as many points from you as it did.

[@CanadianBoardCrew]

Elo systems also have penalties for cheating. Apply those to Jon Jones’s SEVERAL failed drug tests

[@stacrafty]

And Anderson Silva- everyone forgets he was such a notorious cheat he got banned for a year

[@olisk-jy9rz]

Good point!

[@Harinjo916]

They also need to have bonus points for DWP aka Dana White Privilege, on a scale how big of a suck up you are, you get more points.

[@stacrafty]

@Harinjo916 you don't understand how the elo system works. If a fighter gets dana white "privilege" this usually means they have a real test in front of them. Also dana white privilege is more "casual fans don't get it" than actual privilege. Take Izzy getting a title shot over strickland. Casuals forget how Izzy carried the division for 4 years, ducked no one, called people out, never cancelled a fight etc

[@michaelbond7213]

Try figuring out the percentage that a fighter’s ELO represents of the total ELO of their class across their entire career. To correct across time, find the average ELO per class participant in their class during their career and then use that as a factor to scale everyone, which is fair because some classes get more competitive as they have more talent and more veterans. This would also help scale for low participant classes at the high and low weights and early on in the league. This would represent something like a career impact measurement.

[@DRXNESTRIKE]

At that point this man should be getting paid by the UFC 😂

[@nameless5646]

I guess it's damn near impossible to account for accomplishments outside of the ufc but DJ being this low on the goat list feels criminal given that he is on the shortlist of many people in the goat discussion.

[@twnkltoesbrad]

Prob should have weight class elo aswelll w vid tho

[@tuckerjolley7739]

Do a video that shows the average elo of all of a fighter’s opponents that that fighter was victorious against, using the opponent elo at the time they fought. For example, what was the average elo of all of Jon Jones’s opponents, who Jon was victorious against, at the time he fought them, compared to the average elo of all of gsp’s opponents, who gsp was victorious against at the time he fought them? In particular I’m most interested in comparing Conor and Khabib in this way.

[@sprayandpreyy]

I think another important factor to take into account is title fights. Title fight wins should hold a bit more weight than non title fights tbh

[@FerretSnatcher-r2p]

Another idea, you can make a list for every year, so we can see who was the best at the time.