PDA

View Full Version : Pairwise and Bracketology 2013 Edition



Pages : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 [43] 44 45 46 47 48 49 50 51

Priceless
03-21-2013, 08:58 AM
I am off to Boston in a few hours so this will be the last time I am posting from a real computer. Based on the results I expect this is the bracketology I'd put forth.


Manchester (UNH) Providence (Brown) Toledo (BGSU) Grand Rapids (Michigan)
Boston C Quinnipiac Miami Minnesota
Yale Lowell Niagara N Dakota
New Hamp St Cloud Mankato Notre Dame
Union W Michigan Denver Michigan

Now that I've written that, my predictions will start going wrong this afternoon...

Driftryder
03-21-2013, 09:41 AM
The bracket of death would be GR. :eek:

Numbers
03-21-2013, 10:14 AM
Dream bracket for the committee, Priceless. No flights to Manchester, 2 to Providence, 2 to Toledo, 2 to Rapids. I was just thinking about Michigan this morning, too. Wow. What a reward for Minnesota for being #2 overall - playing what would be the hottest team in the country.

80Miracle
03-21-2013, 10:36 AM
I agree this is what the brackets will probably end up (assuming Michigan wins the CCHA Tourney), but I still wish they would spread out the WCHA teams into the FOUR Regions

ExileOnDaytonStreet
03-21-2013, 10:38 AM
A really basic question that I've just realized I never knew the answer to and never inquired about:

How long have the selection committee used the PWR system, and have they ever veered from it for at-large selection or seeding? I've only been paying attention to this kind of detail for the last decade or so, and I can't think of an instance where they've veered from the formula.

Craig P.
03-21-2013, 10:59 AM
A really basic question that I've just realized I never knew the answer to and never inquired about:

How long have the selection committee used the PWR system, and have they ever veered from it for at-large selection or seeding? I've only been paying attention to this kind of detail for the last decade or so, and I can't think of an instance where they've veered from the formula.

They've been using it in full since the 1994-95 season. In 1993-94 (and presumably some years before, but I don't know how many), they mostly used RPI and only used PWR to teams within a small RPI margin of one another on the bubble (i.e. teams within 0.01 or something like that).

Through 1998-ish, if I remember correctly, the committee got the PWR more as a grid of comparisons, and it could be more important for a bubble team to have bubble comparison wins than to have a good overall PWR rank. Somewhere around 1999 or 2000, they appear to have shifted to doing a straight-ranked PWR.

LTsatch
03-21-2013, 12:06 PM
Does anyone know if the Thursday outcomes will have any effect on Yale's possible lock status, I understand that they are right on the edge of being a lock. Thanks for all of the analysis!

burgie12
03-21-2013, 12:11 PM
Does anyone know if the Thursday outcomes will have any effect on Yale's possible lock status, I understand that they are right on the edge of being a lock. Thanks for all of the analysis!
RHamilton's earlier post listed what the updated possibilities would be after each of the possible Thursday outcomes. None of them list Yale as a lock.

Hey all,

I ran all 393,216 outcomes similar to JimDahl, but went a step further and assigned a percent likelihood using KRACH on each individual outcome. I took a brief look, and it looks like my unweighted outcomes match up with JimDahl's

Here are KRACH weighted likelihoods for all seeds:
http://pwr.reillyhamilton.com/

Note that "seed" refers to the PWR seed (after auto-bids have been accounted for), not the PWR rank. Ties broken using RPI.

Individual breakdowns for each of the outcomes tomorrow:

WCHA 'Quarterfinal' 1:

Minnesota State wins: http://pwr.reillyhamilton.com/minnst.html
Wisconsin wins: http://pwr.reillyhamilton.com/wis.html
WCHA 'Quaterfinal' 2:

if Minnesota State won:

Colorado College wins: http://pwr.reillyhamilton.com/minnst-cc.html
North Dakota wins: http://pwr.reillyhamilton.com/minnst-nd.html

If Wisconsin won:

Colorado College wins: http://pwr.reillyhamilton.com/wis-cc.html
North Dakota wins: http://pwr.reillyhamilton.com/wis-nd.html



I'm working on some alternate ways of analyzing the data I've generated. I hope to release a dynamic "what-if" machine after tomorrow's games when the number of outcomes is a bit more manageable and less likely to crash my webhost's database server.

EDIT: Fixed typo per lugnut92 below :)
EDIT 2: Switched domain names.

FlagDUDE08
03-21-2013, 02:07 PM
That seems quite dumb, and how could there be "simultaneous years" -- only one team wins the championship in any year?

I meant consecutive. Oops.

And that's exactly what the NYSPHSAA does. Not very dumb to me.

RHamilton
03-21-2013, 11:46 PM
Up-to-date KRACH weighted breakdown:
http://pwr.reillyhamilton.com/

I just regenerated the KRACH figures, so the weightings are a bit different than the previously posted wis-cc.html file. The unweighted percents and number of outcomes are unchanged.

I'll post some scenarios for the early games tomorrow soon.

Patman
03-21-2013, 11:55 PM
Updated, same model values as inputs.


NAME NCAA S1 S2 S3 S4 CONF ATLRG TUC
1 AA 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
2 AF 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
3 AH 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
4 AI 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
5 Ak 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
6 Ar 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
7 BC 1.0000 0.4229 0.5764 0.0007 0.0000 0.2688 0.7312 1.0000
8 BG 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
9 Bn 0.1201 0.0000 0.0000 0.0000 0.1201 0.1201 0.0000 1.0000
10 BS 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
11 BU 0.1849 0.0000 0.0000 0.0007 0.1842 0.1650 0.0199 1.0000
12 By 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
13 Ca 0.2237 0.0000 0.0000 0.0000 0.2237 0.2237 0.0000 0.0000
14 CC 0.0790 0.0000 0.0000 0.0000 0.0790 0.0790 0.0000 1.0000
15 Cg 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
16 Ck 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
17 Cr 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
18 Ct 0.2615 0.0000 0.0000 0.0000 0.2615 0.2615 0.0000 0.6066
19 Da 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
20 DU 0.9956 0.0000 0.0038 0.8566 0.1352 0.0000 0.9956 1.0000
21 FS 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
22 Ha 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
23 HC 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
24 LS 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
25 MA 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
26 MD 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
27 Me 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
28 Mh 0.1878 0.0000 0.0000 0.0000 0.1878 0.1878 0.0000 0.0000
29 Mi 0.0951 0.0000 0.0000 0.0000 0.0951 0.0951 0.0000 0.3234
30 Mk 0.9993 0.0000 0.0940 0.8962 0.0091 0.0000 0.9993 1.0000
31 ML 1.0000 0.3129 0.6767 0.0104 0.0000 0.3110 0.6890 1.0000
32 Mm 1.0000 0.9033 0.0967 0.0000 0.0000 0.4271 0.5729 1.0000
33 Mn 1.0000 1.0000 0.0000 0.0000 0.0000 0.5355 0.4645 1.0000
34 Mr 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
35 MS 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
36 MT 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
37 ND 1.0000 0.0000 0.5257 0.4743 0.0000 0.0000 1.0000 1.0000
38 NE 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
39 NH 1.0000 0.0000 0.6338 0.3662 0.0000 0.0000 1.0000 1.0000
40 Ni 0.9887 0.0560 0.3854 0.3904 0.1569 0.3270 0.6617 1.0000
41 NM 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
42 NO 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
43 Nt 0.8427 0.1546 0.0470 0.3142 0.3269 0.3404 0.5023 1.0000
44 OS 0.1374 0.0000 0.0000 0.0000 0.1374 0.1374 0.0000 1.0000
45 Pn 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
46 PS 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
47 Pv 0.2562 0.0000 0.0000 0.0006 0.2556 0.2552 0.0010 1.0000
48 Qn 1.0000 1.0000 0.0000 0.0000 0.0000 0.4961 0.5039 1.0000
49 RM 0.0072 0.0000 0.0000 0.0000 0.0072 0.0000 0.0072 1.0000
50 RP 0.1242 0.0000 0.0000 0.0000 0.1242 0.0000 0.1242 1.0000
51 RT 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
52 SC 0.9797 0.0379 0.4296 0.3488 0.1634 0.2286 0.7511 1.0000
53 SH 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
54 SL 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000
55 Un 0.5562 0.0000 0.0000 0.1382 0.4180 0.2518 0.3044 1.0000
56 Vt 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
57 Wi 0.2046 0.0000 0.0000 0.0482 0.1564 0.1569 0.0477 1.0000
58 WM 0.7756 0.0000 0.0000 0.0000 0.7756 0.0000 0.7756 1.0000
59 Ya 0.9805 0.1124 0.5309 0.1545 0.1827 0.1320 0.8485 1.0000

I'll update either VERY late Friday or Saturday morning.

I will also post pairwise code on Saturday with hopes that it will help with future module building.

edit: as in, entice help for the future with the goal of a full on model based monte carlo simulation for use earlier in the season.

RHamilton
03-22-2013, 12:25 AM
I will also post pairwise code on Saturday with hopes that it will help with future module building.

edit: as in, entice help for the future with the goal of a full on model based monte carlo simulation for use earlier in the season.

Cool, will be fun to see others models. Yours is in R? My admittedly sloppy PHP/MySQL database-driven model is here:
https://github.com/hamilr2/NCAA-Hockey-PWR-predictor

Obviously needs a lot of set-up and a DB schema to run, but if you like looking at methodology in an iterative language, it's worth a look. Maybe I'll prepare a readme later.

Team setup and iteration is done in pwr.php, relevant PWR calculations happen in the doPWR function here:
https://github.com/hamilr2/NCAA-Hockey-PWR-predictor/blob/master/include.php#L295

Numbers
03-22-2013, 12:39 AM
I looked at RHamilton's most recent chart. The following schools have a chance yet at the last 2 #1 seeds:
#3 overall: Miami (~55%), BC (~19%), Lowell (~18%), Yale (~1.5%), NoDame (~5%) St Cloud (~1%)
#4 overall: Miami (~33%), BC (~23%), Lowell (~10%), Yale (~15%), Niagara (~6%), St Cloud (~3%), NoDame (~10%).

That is 7 schools still with a chance at a #1 seed, besides the 2 guarantees. Wow. Admittedly, Niagara and St Cloud the smallest, but still.

Numbers
03-22-2013, 12:47 AM
I found an odd thing in JimDahl's blog tonight. The assumption here has been that St Cloud can now clinch with a win.

However, on his blog, updated with today's results, we find this for St Cloud, under the 1 win column:

#13 0.0%

Which in interesting to me, because the #14 row is totally empty. And, if that really means 0.01%, where there might be a strange case yet, then a win tomorrow afternoon does NOT guarantee them in, because we all know that CC, Mich, Prov and Brown could yet win tourneys, in which case #13 in the PWR misses.

Can, RHamilton or PatMan add more information?

Patman
03-22-2013, 12:48 AM
Cool, will be fun to see others models. Yours is in R? My admittedly sloppy PHP/MySQL database-driven model is here:
https://github.com/hamilr2/NCAA-Hockey-PWR-predictor

Obviously needs a lot of set-up and a DB schema to run, but if you like looking at methodology in an iterative language, it's worth a look. Maybe I'll prepare a readme later.

Team setup and iteration is done in pwr.php, relevant PWR calculations happen in the doPWR function here:
https://github.com/hamilr2/NCAA-Hockey-PWR-predictor/blob/master/include.php#L295

right now I'm not running a model I'm using Robin Lock's CHODR model... if I were to run things... and admittedly i operate as a Bayesian, I'd probably do some variant of Poisson regression using the hessian matrix as a measure of parametric uncertainty and then simulate from that assuming normality... I will then get away with any difficult posterior distribution construction and related Markov Chain Monte Carlo inference issues... anyhow...

no, what I will post is straight through, my code to calculate pairwise. I want to build a bigger simulator. I argue that parts would likely have to be written in another language (C or fortran) to achieve desired efficiency... but i seem to say that about most of the stuff i do right now :)

goblue78
03-22-2013, 12:50 AM
I've got PWR code written in Stata if anyone is interested. It's not yet ready to do extensive Monte Carlo simulations yet, because it's w-a-a-a-y too slow on the CommOpp criterion. Everything else is working pretty well, much faster than I'd have expected.

RHamilton
03-22-2013, 01:13 AM
I found an odd thing in JimDahl's blog tonight. The assumption here has been that St Cloud can now clinch with a win.

However, on his blog, updated with today's results, we find this for St Cloud, under the 1 win column:

#13 0.0%

Which in interesting to me, because the #14 row is totally empty. And, if that really means 0.01%, where there might be a strange case yet, then a win tomorrow afternoon does NOT guarantee them in, because we all know that CC, Mich, Prov and Brown could yet win tourneys, in which case #13 in the PWR misses.

Can, RHamilton or PatMan add more information?

I see 10 outcomes (out of 49,152) where St. Cloud ends up #17 when winning their semi.

The scenarios all involve:
- Niagara losing to Mercyhurst or Connecticut
- Miami losing to Ohio State or Notre Dame
- Union DEFEATING Brown in the ECAC Final
- Quinnipiac and Yale TYING in the ECAC Consolation
- 8 of 10 feature BU winning hockey east, the other two have Providence beating BU -- the Providence wins require Connecticut to beat Niagara
- CC winning the WCHA

EDIT: Putting this in the Pairwise Predictor, I see that St. Cloud ends up in a 4-way tie for #9, becomes #12 on tiebreakers. Union finishes #13, and BU finishes #14, but they both move up one seed to make room for Ohio State, Colorado College, and Mercyhurst. This bumps SCSU down to #17.

Ugly is all I can say!

RHamilton
03-22-2013, 03:32 AM
To help all the pairwise prognosticators out there, I put together this:

http://pwr.reillyhamilton.com/query.php
Simply input your choice, or fill in games as the day goes on, and receive the weighted (and unweighted) outcomes.

It may take up to a minute or so for some queries. The more teams you switch from "Don't Care" to something else, the quick it will run. Impossible conditions (a team winning a final than won neither semi) will not return any results. Results are cached, though, so repeated visits to the same query should be instantaneous. Use the links to share.

Let me know if you have any issues running it. We'll see how long it stays up, and if my webhost complains!

Umileated
03-22-2013, 07:44 AM
Maybe I'll prepare a readme later.

Yes, please!

sshablak
03-22-2013, 09:11 AM
Much easier....you're good. Tx