USCHO Fan Forum
This is a sample guest message. Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!
-
The USCHO Fan Forum has migrated to a new plaform, xenForo. Most of the function of the forum should work in familiar ways. Please note that you can switch between light and dark modes by clicking on the gear icon in the upper right of the main menu bar. We are hoping that this new platform will prove to be faster and more reliable. Please feel free to explore its features.
Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
- Thread starter Umileated
- Start date
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
I predict that any of that will only come from one site.
I predict that any of that will only come from one site.
Possibly the first ever appearance of "NCAA" and "integrity" in the same sentence?
MavHockey14
New member
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
Thank you to everyone who put in all of the time and effort into trying to predict the field and understand the Pairwise. Seriously. Greatly appreciate it.
Thank you to everyone who put in all of the time and effort into trying to predict the field and understand the Pairwise. Seriously. Greatly appreciate it.
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
Plus one on that, great effort and problem solving well above my level.
Plus one on that, great effort and problem solving well above my level.
P
Priceless
Guest
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
And since that site doesn't have a whole lot to brag about, problem solved.
UnionHockeyManiac
Union Hockey fan for life
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
Yes, a BIG thank you to everyone in this thread (Priceless, Numbers, RHamilton) who provided information and answered every single question. Much appreciated!
Yes, a BIG thank you to everyone in this thread (Priceless, Numbers, RHamilton) who provided information and answered every single question. Much appreciated!
Patman
Rodent of Unusual Size
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
Does anybody have the results in jtw's format? Maybe even more years on top of that... I think its time I did a bit of sensitivity analysis (of sorts) on the data set. Since I have the entire pairwise I should be able to find out what happens if I delete 'X' number of results.
I suppose to keep the NCAA from being mondo ****y about "we evaluate at the end of the season" I can always void the result and re-run but that might take a bit of model building.
Both are up for grabs in this context, I suppose. However, if it is the latter then I'll need more than just jtw's information (home/road/neutral info).
Does anybody have the results in jtw's format? Maybe even more years on top of that... I think its time I did a bit of sensitivity analysis (of sorts) on the data set. Since I have the entire pairwise I should be able to find out what happens if I delete 'X' number of results.
I suppose to keep the NCAA from being mondo ****y about "we evaluate at the end of the season" I can always void the result and re-run but that might take a bit of model building.
Both are up for grabs in this context, I suppose. However, if it is the latter then I'll need more than just jtw's information (home/road/neutral info).
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
PatMan,
This will probably seem foolish to you, as I am no programmer and don't know what you mean.
But, if you want the results in jtw's format, and the results are the game results from this year, can't you just copy them off of his site?
PatMan,
This will probably seem foolish to you, as I am no programmer and don't know what you mean.
But, if you want the results in jtw's format, and the results are the game results from this year, can't you just copy them off of his site?
Patman
Rodent of Unusual Size
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
probably... but I'm being lazy
probably... but I'm being lazy
Patman
Rodent of Unusual Size
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
Between a set of planes, rental cars, beers, and everything else I've started to do some analysis of PWR vs. KRACH (KRACH as I needed a baseline).
I'm still doing some initial analysis... pain in the *** to construct. I've ran what is called a "bootstrap". Its a re-sampling procedure meant to acquire the features of statistical results... variation and the rest.
I looked at the 1-59 absolute difference. This is just a start. I add up the difference in the ranks between a randomly drawn re-sample from the results of the entire season* and the season as it actually occurred. For each random "season" I ran both PWR/RPI and KRACH (modified for 0 win or 0 loss situations). I got the following...
I should note that the pairwise/RPI has some high outliers... it would appear from this season's data (and I wouldn't draw conclusions without several years of analysis) that KRACH is more stable under this metric from 1-59.
Other things to look at later... square of the differences and finding good metrics to refine down to a target (top 16, whatever else.)
*A bootstrap is a random re-sample under some very loose assumptions. The nature of the re-sampling implies that games between two opponents is a random event... that is the teams are random. We know this is wrong and is in fact highly dependent and often fixed. There are other variants... which would take longer... this is still exploratory.
edit: I could probably run a frequentist test on this... yes... but i don't care about that right now. I couldn't really say what the significance of the above truly is. Every method will produce a suitable amount of error in the presence in so few events.
Between a set of planes, rental cars, beers, and everything else I've started to do some analysis of PWR vs. KRACH (KRACH as I needed a baseline).
I'm still doing some initial analysis... pain in the *** to construct. I've ran what is called a "bootstrap". Its a re-sampling procedure meant to acquire the features of statistical results... variation and the rest.
I looked at the 1-59 absolute difference. This is just a start. I add up the difference in the ranks between a randomly drawn re-sample from the results of the entire season* and the season as it actually occurred. For each random "season" I ran both PWR/RPI and KRACH (modified for 0 win or 0 loss situations). I got the following...
Code:
EVAL: SUM OF ABS DIFF OF RANKS
10,000 BOOTSTRAP RE-SAMPLES
PAIRWISE EVAL BY QUANTILES
2.5% 5% 10% 25% 50% 75% 90% 95% 97.5%
328 344 362 390 424 460 498 522 552
KRACH EVAL BY QUANTILES
2.5% 5% 10% 25% 50% 75% 90% 95% 97.5%
316 330 346 374 406 444 478 498 518I should note that the pairwise/RPI has some high outliers... it would appear from this season's data (and I wouldn't draw conclusions without several years of analysis) that KRACH is more stable under this metric from 1-59.
Other things to look at later... square of the differences and finding good metrics to refine down to a target (top 16, whatever else.)
*A bootstrap is a random re-sample under some very loose assumptions. The nature of the re-sampling implies that games between two opponents is a random event... that is the teams are random. We know this is wrong and is in fact highly dependent and often fixed. There are other variants... which would take longer... this is still exploratory.
edit: I could probably run a frequentist test on this... yes... but i don't care about that right now. I couldn't really say what the significance of the above truly is. Every method will produce a suitable amount of error in the presence in so few events.
Last edited:
Patman -- you and I are definitely going to have to get together in Pittsburgh to talk modeling... Though definitely after Thursday afternoon.
Patman
Rodent of Unusual Size
We'll see... I'm hoping to get something going at Friday but it would relate to work through my employer... If I can snare something to do at CMU I'll only have to torch 2 vaca days
Patman
Rodent of Unusual Size
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
If you guys want it, I am talking to a CMU professor and assoc editor of JQAS to have a math/stat discussion on sports stats... Basically meet at the bar and shoot **** in regards to things, numbers, analytics, hockey, and hopefully all three at once. This won't count as work for me but if we can do that I think that'd be fun and helpful for those interested.
If you guys want it, I am talking to a CMU professor and assoc editor of JQAS to have a math/stat discussion on sports stats... Basically meet at the bar and shoot **** in regards to things, numbers, analytics, hockey, and hopefully all three at once. This won't count as work for me but if we can do that I think that'd be fun and helpful for those interested.
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
You don't work for ERS do you?
Patman
Rodent of Unusual Size
I wok for the gov't
Edit: no, I don't work for agriculture, I'm in commerce, check your email
Last edited:
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
That doesn't sound like simple math.
That doesn't sound like simple math.
Re: Pairwise Bracketology Part Deux: Playing with RHamilton's Tool
If I can make it fit into whatever else may be going on, I'm there.
If I can make it fit into whatever else may be going on, I'm there.