2019 USCHO Pairwise Predictor

[QUAlum2004]

Re: 2019 USCHO Pairwise Predictor

Season not officially over for Union. CHN found one scenario (actually 4) that gets Union into the tournament. Its unlikely but anything is possible. It doesn't matter what the Atlantic Hockey teams do by the way because if any of the 4 win then Union would still be in.

https://www.collegehockeynews.com/ra...0#.XJKb9ChKiUm

[ThePowerConference]

Re: 2019 USCHO Pairwise Predictor

Everything else I keep reading from other sources says that St Cloud is the guaranteed top overall seed based on any and all outcomes this weekend. But for some reason the Exiled One can still find a route that shows Mankato passing them. Maybe that "bug" has now been fixed with the home/neutral sites for the Big Ten and WCHA, per Jim Dahl.

[JimDahl]

Re: 2019 USCHO Pairwise Predictor

After working with the CHN guys a bit on it, their calculator is now producing the same results as my forecasts. It was the home/neutral issue, and they're now treating Big Ten and WCHA championships as home games for the hosts. Now that I trust the results again, I'll see if I can whip up some interesting analysis in the remaining 2 days.

Here's the most unlikely one:

Union gets in!

Obviously you need all but Atlantic Hockey to be won by teams that otherwise finish in the top 15. Then you have to get Union up to 15, which is a delicate balancing act (e.g. the scenario above pushes Union's RPI .0002 above Notre Dame's); so many scenarios that meet the previous criterion, like Mass defeating Northeastern, wouldn't work. Of about 49,000 possible remaining outcomes, Union makes it in 20.

(Of course, due to all the issues discussed in this thread, that scenario won't produce the same result in USCHO's calculator).

[The Exiled One]

Re: 2019 USCHO Pairwise Predictor

I've been poking at it for about a day, and still don't feel that all the issues are worked out. The calculators all give different results in important edge cases.

I'm not yet worried about the committee's calculations. Calculating PWR from a list of games is relatively easy, and I haven't seen a different result out of our three implementations (CHN, USCHO, and my own) since the last change in formula. So far, these errors are consistent with problems with the user interfaces that the sites use to allow users to specify some of the results in the lists of games. But, until we figure it out, that's just my best guess.

I'm not sure the USCHO is even trying to fix their predictor tool, which appears to be miscalculating the common opponents variable.



For the Sum of Win% column, I believe it should read 1.000 for both teams. Somebody correct me if I'm wrong.

[JimDahl]

Re: 2019 USCHO Pairwise Predictor

I've been poking at it for about a day, and still don't feel that all the issues are worked out. The calculators all give different results in important edge cases.

I'm not yet worried about the committee's calculations. Calculating PWR from a list of games is relatively easy, and I haven't seen a different result out of our three implementations (CHN, USCHO, and my own) since the last change in formula. So far, these errors are consistent with problems with the user interfaces that the sites use to allow users to specify some of the results in the lists of games. But, until we figure it out, that's just my best guess.

[TheRevengeance]

Re: 2019 USCHO Pairwise Predictor

NU shouldn't be able to pass St. Cloud as far as I know. The current common opponents are Union (both teams have a .000) and BC (SCSU had a 1.000, NU currently has a .667). SCSU did not play anybody else on NU's schedule.

[J.D.]

Reading these thread makes one hope that whatever software the committee uses to pick the teams is bug free.

Was thinking the same thing

[Ralph Baer]

Re: 2019 USCHO Pairwise Predictor

Reading these thread makes one hope that whatever software the committee uses to pick the teams is bug free.

[JimDahl]

Click on the link in my post. I think the USCHO predictor is broken. I don't think Northeastern can win the common opponent comparison.

If you keep track of the records of the teams, the tie seems to propagate strangely, e.g. SCSU's record increases by multiple ties in some scenarios in which I give them a tie. Of course, it's possible that's a display bug, not a calculation bug, but I can't consistently duplicate those scenario results from USCHO, either.

I can consistently duplicate CHN's results by being willing to switch back and forth between counting the on campus championship games as home or neutral, but which I have to choose depends on the other set of games, and there's no obvious pattern. Makes me wonder about some sort of cache, but I can't figure it out.

I suppose either could be an issue of previous runs not completely resetting, but I worked pretty hard to start clean on my tests on CHN.

[The Exiled One]

Re: 2019 USCHO Pairwise Predictor

Northeastern beat St Cloud during the regular season so they would win the head to head comparison, right?

I can't find a scenario that puts Mankato ahead of St Cloud as the top overall seed.

Click on the link in my post. I think the USCHO predictor is broken. I don't think Northeastern can win the common opponent comparison.

[ThePowerConference]

Re: 2019 USCHO Pairwise Predictor

Minnesota State #1 overall (meaning they'd be placed in Fargo)

EDIT: Can someone explain to me how Northeastern can win the common opponent comparison with SCSU?

Northeastern beat St Cloud during the regular season so they would win the head to head comparison, right?

I can't find a scenario that puts Mankato ahead of St Cloud as the top overall seed.

[The Exiled One]

Re: 2019 USCHO Pairwise Predictor

Minnesota State #1 overall (meaning they'd be placed in Fargo)

EDIT: Can someone explain to me how Northeastern can win the common opponent comparison with SCSU?

[klbaum1077]

I got BG in unless: BU/'BC win out, or CC wins out. Otherwise I can't find a way BG is out.

What if brown wins ecac and bg loses

[The Exiled One]

Re: 2019 USCHO Pairwise Predictor

Weird. When I do the same thing on CHN, seems like SCSU is still number 1

Not only that, but if you look at the tables for that scenario it shows a different RPI and St. Cloud still wins 58 comparisons. Something is broken there.

EDIT: Never mind, figured it out. ticapnews forget to put in a result for the NCHC third place game. SCSU can lose comparisons to Northeastern and Union.

[TheRevengeance]

Jim, I tweeted at CHN asking and it appears they intended for these to be calculated as home games but it isn't happening for some reason. So it looks like you're right.

Update: CHN seems to have updated their calculator to have the home/road weights applied. Uscho and CHN now give me different RPI results for those games. If you're still seeing the same results on CHN maybe clear your browser.