We have the hardest remaining schedule in CFB

FPI blows. Maybe if we all just stop talking about ESPN will make it go away.
 
Arkansas still #19 tells you all you need to know.

But of course us at #1 in most anything is always cool so that part is probably correct.
 
Our projected W-L is 8.7 - 3.8. For those of you counting, that's 12.5 games. Does compute, apparently, but I don't get it.
 
It's not a bad idea to try to develop a model like this. But it's silly for them to keep marketing it so hard when it's giving ridiculous results. I hope somebody is keeping track of the results so we can analyze it's accuracy in picking games/spreads.

I don't know all the inner-workings, but just from looking at the output it seems pretty clear that it puts too much weight on recruiting rankings and not enough on recent history / coaching.
 
ThisIsAtlanta said:
Our projected W-L is 8.7 - 3.8. For those of you counting, that's 12.5 games. Does compute, apparently, but I don't get it.
Click to expand...

It includes the possibility of playing in the ACCCG I believe. Teams that are much less likely to win their division (and Big 12 teams) are closer to or at 12.
 
Ohio State. 40% chance of winning out, SOS rank >60.
 
I would love a chance to nerd up and make my own predictor, but I have a kid on the way and a nursery to build in my off hours. It might be an offseason endeavor.

TIA - where do you get a database for your predictor? Is it free?
 
I'm sure the weightings for recruiting rankings don't come out of thin air, but with some kind of correlation. I don't think it's just the previous year's rankings either. You also have sophomores replacing seniors. I would use the sophomore's recruiting ranking for lack of anything better.
 
I would love for whomever created the logic behind the FPI to explain how Tennessee at 1-1 is ranked 12th
 
G776T said:
I would love for whomever created the logic behind the FPI to explain how Tennessee at 1-1 is ranked 12th
Click to expand...

because their only loss is to Oklahoma in overtime? And Bowling Green looks pretty legit right now?

Plus the recruiting weighting factor which in my opinion is a bit too strong in ESPN's FPI ranks. That's why we're consistently undersold, which isn't good considering the CFP committee gets its stats from ESPN and its research arm. FPI is definitely part of the consideration as Game Control (unique to FPI) was mentioned as part of the decision-making process.
 
G776T said:
I would love for whomever created the logic behind the FPI to explain how Tennessee at 1-1 is ranked 12th
Click to expand...


They're only loss is to the same team that beat the #12 ranked team. Pretty impressive.
 
Hopefully FSU doesn't get exposed before we play them.

I have a feeling UGA definitely will. Their offense is a dumpster fire.
 
Arkansas also outgained Toledo by 200 yards. Pundits hate talk of moral victories and put all weight on wins and losses. But computer rankings have to predict FUTURE contests.

For example, the equivalent of FPI for basketball gave Kentucky a <50% chance to win the championship. That's when most pundits assumed Kentucky would easily run away with it, but the model saw how many of Kentucky's wins were questionable.
 
Flywheel said:
I would love a chance to nerd up and make my own predictor, but I have a kid on the way and a nursery to build in my off hours. It might be an offseason endeavor.

TIA - where do you get a database for your predictor? Is it free?
Click to expand...

I use a webscraper to obtain data from ESPN's game pages, the NCAA's game pages, and Yahoo's recruiting site. I've been collecting and archiving data from various sources since 2011, and have managed to get a pretty comprehensive database going back to 2002 including an archive of all of the old cfbstats data.

It's free aside from the small amount of time I spend retooling the webscraper every time ESPN, NCAA, or Yahoo updates their site format. The data is not really in a good format for public consumption, however. I pull most of it down in JSON and store it that way. Since python can build native objects from JSON and serialize them back, I also work with it in that format, so it no longer enters a database or even a spreadsheet until I have a specific need to analyze it in a different format. That hasn't happened in a while.

If you're interested in free data, it means work. If you're interested in fast, timely, complete, well formatted data, you may end up having to pay for it. The last free reliable source I used was cfbstats, which has since become SportsSource analytics. They'll run you a couple hundred bucks per season for data. There was an open source python ESPN webscraper on github for a while that looked like it was working, but the last time I checked it was before ESPN updated their page format about a week before kickoff this year. No idea if it's still ticking. There are some football nerd types on some of the SBnation blogs that also appear to have some sort of advanced data source, but I have never bothered to research it. I'd be interested to know, though, if anybody else knows more about that.

We also have a couple of stats gurus on this site aside from myself who might know of something.
 
I spent some time last night working on a new predictor. It's gonna murder TiA's.
 
You must log in or register to reply here.
Back
Top