How Madden Ratings Are Made | Hacker News Reader

How Madden Ratings Are Made

fivethirtyeight.com

46 pointslt14 comments

> “I’ve heard it speculated that even the Madden team themselves don’t really know what’s going on,” Bailey said. “Because they’ve got years upon years of systems and code just layered on top of one another, where it’s not always entirely clear how they’re interacting.

It's been several years since I worked on Madden, but that was certainly true at the time. It was like an archeology site with layers of new code plastered on top of older stuff with at on top of older stuff. Tons of dead code, tons of undocumented code where the original author had moved on and no one knew how it worked or what it did.

Having a 9-month dev cycle and an upper management structure that doesn't know much about software does that. Every cycle it was, "we've got less than a year to get a list of bullet points on the back of the box implemented". Anything like architecture or refactoring was the first to get cut.

There was certainly good code in there too, but, man, it really needed cleaning up.

Probably doesn't help that the ratings guy seems to constantly be pushing for more ratings to be added, when it's not even clear what the existing ones do.

I've had the same problem when making http://basketball-gm.com/ even though I'm just one person and one of my design decisions has been to keep it as simple as possible for precisely this reason. Large models with lots of parameters are hard, no way around it.

I agree in general, but how do you capture all the nuance w/o more models. Sherman and Revis are the best two corners in the game (arguably), but super different styles of defense.

Another example is Russell Wilson. In the pocket he's a good passer, on the run he's a great passer.

People who are fans of these sports want to see these relatively small differences. I'm not sure how else you do that w/o a lot of attributes.

It's a balance, but the answer isn't always to add more parameters. Like, just because you have a parameter called PassingAbilityWhileRunning doesn't mean it will actually reflect passing ability while running if your model is too complex to understand. Maybe that's a bad example, but you get the idea. An attempt to add nuance can instead add confusion and make the game worse. But like you said, you undoubtedly end up with a fuckton of parameters if you want to have a somewhat realistic game.

There's no perfect answer, but it is definitely a good thing to resist adding parameters as much as reasonably possible.

This is a bit off topic, but something I wanted to know for a long time.

How does match playing simulation work in sports games? For example, given Yankees (strength: 89/100) vs Red Sox (strength: 88/100), how does a program come with with score like 5:2, Yankees win or 3:7, Red Sox win?

Even just a name of the algorithm I can look up would be super helpful.

It depends on how you scale the strength ratings, a one point advantage in strength could mean a .75 win probability or a .51 win probability. Here is how they handle it in the Elo rating system:

http://en.wikipedia.org/wiki/Elo_rating_system#Mathematical_...

Modern sports games usually need the same stats from a simulated game as they would get from a played game so simulating is often just running a stripped down version of a full game.

Something like Football Manager works as a genuine agent-based simulation of soccer, but there's also a quick simulation which is probably just based on statistical models.

The biggest problem always seems to be human intervention and the expectation of how much and where a human might intervene in any play.

In addition, there are global sliders which have some kind of effect on performance. Whether these are just offsetting player attributes or introducing some other factor I'm not sure.

My personal gripe with the latest offering is that recievers most of the time will make no serious effort to catch a ball. They are on-rails to a catch point and do little else.

Not only was this article interesting, but the website was great. The informative bits that dynamically appear while scrolling + the interactive rating system was very intuitive and smooth. Very impressive.

It broke the back button severely.

The experience on my mobile phone was awful. The text was jittering horribly every time I scrolled. I finally gave up and opened the article on my computer.

It's amazing that a big studio like EA has allowed a single developer to maintain such a stranglehold over one of their largest franchises for so long. It makes you wonder how often they quietly call meetings behind his back that are solely focused on replacing the mad scientist.

I don't think Donny is the only one who has worked on player stats over the years. This wasn't at all the side of things I was on when I worked on Madden, but I can think of at least one other person who also hacked on the player DB.

It's not clear that he has a "stranglehold". Sure, he does a singular, and important job. But that doesn't necessarily mean he isn't replaceable at that job.

Reply on news.ycombinator.com