News Article: Interesting computer rankings
- Thread starter Bamaro
- Start date
Garbage in = garbage out.Alabama at 10th...?
I think I know what happened to that fifteen year old from awbern...
They do have quality wins over Delaware, Tulane, Indiana, and Duke, which beat Notre Dame at home!Wake Forrest over us!
Last edited by a moderator:
Anytime Wake Forest is on a Top 10 Football List -- even if it's week 1.....the software must be checked for a virus.....
True. I saw a preseason a big SEC publication that tried to predict every score of every team for entire season lol. First poll after week 2 imo.
I have kept up with the computer rankings for years and the Colley is a joke. Billingsley is not in the top ten either. If you are interested here are the top four I use to help in pickems and the record thus far:
Atomic: 41-11
Basset: 36-16
Sagaran: 35-17
Vegas Odds: 35-17
Massey: 34-17
There are a couple more, like Pi-rate and Ashby are rated at the top, like atomic, but Pi-rate is too hard for me to figure out and Ashby comes out late in the week, so I dropped them. Massey has a list of them if you want to explore.
http://masseyratings.com/cf/compare.htm
Atomic: 41-11
Basset: 36-16
Sagaran: 35-17
Vegas Odds: 35-17
Massey: 34-17
There are a couple more, like Pi-rate and Ashby are rated at the top, like atomic, but Pi-rate is too hard for me to figure out and Ashby comes out late in the week, so I dropped them. Massey has a list of them if you want to explore.
http://masseyratings.com/cf/compare.htm
Tennessee goes on the road for the first time the next two games against UGA and TAMU. TAMU gets a warm up game on the road at USC before coming home to play the Viles.
The next two weeks should tell us a lot about both.
The next two weeks should tell us a lot about both.
Well, when the result of half the differential equations is NaN, you're going to have a lot of screwy results.
Colley - as well as Billingsley, which another poster mentioned - seems to be extremely reliant on relational comparisons, which makes them reliable only once a significant number of games have been played. From my vantage point, Colley is really only good at being a predictive model a week or maybe two before the conference championship games are played.
Full disclosure, I know the guy who runs Atomic. I'd like to pass along things he's told me about rankings:
- the old BCBS rankings were *supposed* to be W/L only, no scores, which amounts to about 11 bits of information on each team for a whole season
- his method is Bayesian, which he describes as "knob-less", i.e. there are no knobs you can tweak, all the numbers come out naturally directly from the math
- the best ranking/chance of winning comes from the Vegas line, not from computers
- many computer rankings "cheat" by salting in some of that Vegas line into their numbers
- you can correlate the computer rankings with Vegas, and figure out what that percentage cheat is, i.e. you can more-or-less reverse engineer each computer algorithm at the end of the season. Basically, the more you cheat, the more accurate your "computer" ranking is.
He told me it is quite a bit of work, keeping up with all the scores and entering them, for hundreds of games per week.
- the old BCBS rankings were *supposed* to be W/L only, no scores, which amounts to about 11 bits of information on each team for a whole season
- his method is Bayesian, which he describes as "knob-less", i.e. there are no knobs you can tweak, all the numbers come out naturally directly from the math
- the best ranking/chance of winning comes from the Vegas line, not from computers
- many computer rankings "cheat" by salting in some of that Vegas line into their numbers
- you can correlate the computer rankings with Vegas, and figure out what that percentage cheat is, i.e. you can more-or-less reverse engineer each computer algorithm at the end of the season. Basically, the more you cheat, the more accurate your "computer" ranking is.
He told me it is quite a bit of work, keeping up with all the scores and entering them, for hundreds of games per week.
Scores are nothing compared to the results of every play they have over at cfbstats.com
If ever you want to develop analytics, they have the data for you, ready to download.
It used to be free, but now they charge.
Thanks. I know it takes a lot of work to enter the data for the computers to crunch but I had no idea of how it worked.
I have found that stats work well picking between Power 5 teams playing Power 5 teams. But put a Power 5 vs a Group of 5 and the stats don't mean very much. I am trying to devise a stat system to account for that. You know, I keep tweeking and tweeking trying to find that system that is very accurate.
Oh, BTW Atomic is better than Vegas so far this year on straight up. Now this is only on games under a TD on odds and only the teams on the four different pickems I am on.
Last edited:
Interesting. No idea there were so many.
The best hidden gem: one has us ranked #15, which (by quick perusal) is the lowest. Might be karma, but they are abbreviated as "LSD", at the top of the chart.
The only way it could have been funnier is if they made it LSU.
Latest threads
-
-
Question: Chris Youngblood did he commit over the weekend?
- Started by mlingerfelt31
- Replies: 3
-
-
-
-