Fight Night Champion: Converting Ratings From Title Bout

I have begun an endeavor to attempt to fix the solid but flawed Fight Night Champion video game, adjusting fighter ratings based on the classic Title Bout game, as while I love Title Bout boxing just begs for graphic representation. EA supposedly has a new version in the works, but it's taken a back seat to their UFC franchise, and this ESports Boxing Club game in development has promise, but the lack of real details let alone a release date makes me think we're a year away at least from seeing that come to fruition.

My conversions are solid for fighters who were pretty standard boxers or sluggers. The game's biggest limitation to me is in handling - or not handling at all - a defensive fighter. Getting a Pernell Whitaker, Willie Pep or Nicolino Locche to fight as they did is the biggest chore, because the game doesn't handle defense great.

So for example, my Barrera v Morales tests looked solid. But pitting Meldrick Taylor against Locche has been irritating, because Locche in particular may as well have been named John Smith through many of these sims.

But this latest one shows promise.

Locche takes decision, 97-93, 97-93, 96-94. Taylor pushes the action, outpunching Locche 713 to 384, but Taylor lands only 21% compared with 34% for Locche.

In early stages of doing this, so there will be lots more tinkering, but this has been the biggest issue I've faced in the early going and I'm comfortable in at least making good progress.

I'll keep any updates on this in this thread so as not to spam the board with it.

GH

 

Some test bouts run today:

Jeff Fenech vs Azumah Nelson: Majority Draw 12, 115-113 Nelson, 114-114 twice

Vernon Forest vs Paul Williams: Forrest UD 10 97-93, 97-93, 96-94

Mike McCallum vs Julian Jackson: McCallum dominates before KO 8

Erik Morales vs Marco Antonio Barrera: Morales SD 12 115-112, 114-113, 113-114

Julian Jackson vs Nino Benvenuti: Jackson KO11 (i don’t mind the KO here as Jackson could do so to pretty much anyone, but he manhandled Benvenuti all fight, so I may have Nino a bit underrated here)

 

I just ran Hector Camacho vs Ricky Hatton, 10 rounds.

In several tests as I tweaked the fighters, Camacho would control early, Hatton would wear him down late and pick up the knockout.

But Camacho was never knocked out in his career, so while I could see Hatton wearing him down as the fight went on, I didn't feel it was right having Camacho get stopped, especially at 10 rounds.

Last sim had Camacho sweep the first 7, Hatton takes the last three rounds as Camacho starts to feel the effects. Over 12 or 15, Hatton probably wears him out.

 

Doing a guy from Long Island, my neck of the woods, Howard Davis Jr.
I kept having him defeat Nicolino Locche, which even at Davis' peak I don't see happening. Took a lot of tweaking but finally got a balance that has him fighting relatively as he did but not good enough to stop Locche, who took a unanimous decision in the last sim.

 

Latest test…Wilfred Benitez rallies from a second-round knockdown to take a unanimous decision from Vernon Forrest!

 

Lots of work on this over the past month. Happy on the whole, but having a devil of a time striking the balance between aggressiveness, punch output and stamina. Getting closer on it but easily the most frustrating part of trying to get this done. The speed at which guys who throw a good number of power shots lose their stamina in the game is too fast and leads to them getting knocked out by guys who shouldn't really knock them out because they become susceptible when their stamina drops.

I am however confident the finished product here is going to be really solid. Just takes some effort and tweaking!

 

I have been working over the past month or so on creating a system to convert ratings from the classic Title Bout boxing game to fighter ratings for the Fight Night Champion video game, which is easily the best graphical representation of the sport ever done, though the game is 10 years old.

There has been lots of testing and re-testing, but I am feeling like I am close to something that works, at least as best as possible within the confines of the game.

One-off fights are good, and seeing things go as I want in one-fight scenarios is great and has led to lots of progress. You can see photos fro some of those on the right.



Now the hard part begins.

I have between 20-25 fighters per weight class set up to evaluate (for now). I have most of their base ratings developed, now comes the tweaking. To tweak them, I need more head to head matchups.

So here's the plan: I'm going to put together fighter "pods" of 6 in a weight class at a time.

1. Run 100 sims in Title Bout. Results that occur at least 20% of the time will be considered an "acceptable" result to occur in Fight Night.
   
   
2. Each fighter in the pod will fight each other fighter in the pod once in Fight Night.
   
   
3. Both fighters will get points for a result in line with the Title Bout simulations. The top result is "Ideal" and gets 10 accuracy points. The next acceptable result is worth 9 points if it happened within 5 times of the ideal result in the Title Bout sims, 7 points if it happened more than 5 times fewer than the ideal. The next acceptable result is worth 5 points if it happened within 10 times of the Ideal result, 3 points if it happened more than 10 times fewer than the ideal. A result in Fight Night that occurred less than 20% of the time in Title Bout gets both fighters 0 points. Ideal results also get bonus points as follows: If a KO is the same round as the average stoppage round in Title Bout, it's 3 bonus points. If it's within 1 round, it's 1 bonus point. An ideal decision that matches the type of decision in Title Bout (usually unanimous) also gets 1 bonus point.
   
   
4. After all pod fights are over, a fighter will have a score I'll call a "Confidence" score - as in the confidence I have this fighter is rated correctly. This is the percentage of the 50 "Ideal" points that could have been attained that a fighter got. A score of 70% is passing and will be good enough for me to put a first version out there (obviously to be tweaked over time). A score of 100% of course is full confidence the fighter is rated appropriately.
   
   
5. If less than 3 fighters in a pod pass, the whole pod will be restarted and adjustments made as needed. After all, if more than half the fighters in the pod need to be adjusted, I should probably be running everyone through against everyone else again. But if, say, 4 fighters pass and 2 fail, those 2 will just be entered into a different pod to undergo the process again.
   
   The hope is that doing these pods gives me a better idea of how the fighters are really built instead of just getting the "right" result in one fight against one opponent. And, the more of these pods I do, the more I'll see "hey, my formulas seem to be missing the mark mostly on aggressiveness or whatever" which will make fixes quicker and easier the more tests I run.
   

 

My plan was always to use my formulas to set the framework of the fighter, the base "ratings" with the proper results of who beats who, etc., in version 1, then address style in V2.

But I decided to start poking around more on style stuff, particularly focused on punch output. If you have a guy who throws 65 punches a round in real life and I turn him out throwing 35, the fight result being right doesn't matter because it won't FEEL right.

And so I started working with CompuBox numbers. I created a dummy "Joe Average" fighter to run guys against, and started playing with their aggressiveness and punch frequency ratings a bit to try and get them representing themselves in better fashion.

And, in very early returns, this looks REALLY good

I'm a bit sick of watching Meldrick Taylor and Wilfred Benitez at this point but they're my test subjects. There's not a LOT of punch stats available on them, but there's enough to work with. Before 1985 there are really only data compiled for big fights, and not all. So I'll have to work a little harder on those guys. But I digress.

HERE is a chart for Taylor and Benitez (click the image to expand it). On the left are data from three fights for Taylor and two for Benitez (the only two they had for him, but his two best known, vs Leonard and vs Duran). You can see the averages of that data next to the left columns. On the right are the results of three three-round sims in Fight Night Champion.

Fighter stats can vary greatly from fight to fight. For example, Timothy Bradley's three chosen fights, his punches per round range from 43 to 62 to 89!

So I set my allowance at 20% to be acceptable, meaning punches thrown, jabs thrown, power punches thrown, punch landed % and punches landed against % need to fall within 20% in the sim of their real-life range. I'll settle for 4 of 5 if I have to, but ideally I obviously want 5 of 5.

For Benitez and Taylor, both guys came in within range for ALL 5 categories!

I've now decided to flip the script and focus on style FIRST, then work the other ratings around that. My formulas generate a range for the key ratings anyway, so adjusting within those ranges will be fairly easy (famous last words). But having fighters perform as they really do/did is essential to get right, and I now have a far greater belief I can do that accurately!

 

cool

 

ok