PDA

View Full Version : 2014-2015 D1 "GRaNT" Computer Rankings



Pages : [1] 2

TonyTheTiger20
01-25-2015, 04:06 PM
More fun with Excel spreadsheets!

I created my own computer ranking last night, called the GRaNT Computer Rankings -- (Stands for "Grant's Reasonable and Not Terrible Computer Rankings"). Pretty exciting stuff. here is the link to the ranking. (http://www.bcinterruption.com/boston-college-hockey/2015/1/28/7925221/introducing-the-mens-womens-hockey-grant-computer-rankings-ncaa-frozen-four).

Here's the jist of the rankings:

-All teams are assigned an arbitrary rating. It doesn't matter what you use because the rankings will come out the same regardless of what you start with. Since it doesn't matter, the calculator literally starts each team out with a RAND() value.

-Each team's rating is recalculated based on their winning percentage and their opponent's rating as follows:
---A win adds the opponent's rating
---A loss adds zero
---A tie adds 1/2 the opponent's rating
------The sum is divided by the number of games played, and then adjusted to equal the percent of the average of the new ratings.

The GRaNT Rankings are constructed somewhat similarly to KRACH in that it uses each team's ranking to calculate everyone else's ranking. I'll explain.

Example: Teams A, B, and C each play each other once each.

Team A's starting rating: 0.666
Team B's starting rating: 0.444
Team C's starting rating: 0.222

Team A beats Team B and ties Team C. Team A's new unadjusted rating is (0.444+0.111)/2=0.278.

Team B beats Team C and loses to Team A. Team B's new unadjusted rating is (0.222+0)/2=0.111

Team C ties Team A and loses to Team B. Team C's new unadjusted rating is (0.333+0)/2=0.167

The average of the new ratings is (0.278+0.111+0.167)/3=0.185

So, each team's new rating is:

Team A: 0.278/0.185=1.503
Team B: 0.111/0.185=0.600
Team C: 0.167/0.185=0.902

Then the process repeats itself, recalculating based on the new ratings instead of the starting rating. Each team's rating will converge to a different number such that when you recalculate, you get the same number:

Team A: 0.6190/0.4491=1.3784 (percent of average)
Team B: 0.3837/0.4491=0.8544
Team C: 0.3446/0.4491=0.7673

Team A: (0.8544+0.38365)/2=0.6190
Team B: (0+0.7673)/2=0.3837
Team C: (0.6892+0)/2=0.3446

-- and those convergent values are each team's final rating.

That final rating is scaled to make zero equal average, and ranked, naturally, from best to worst. The result is the GRaNT Rankings.

You'll notice that teams are awarded 0 points for a loss, no matter who they play -- that means that teams are not penalized for losing to good teams, since all losses count the same. The end result is that teams are rewarded for playing and beating good teams, not penalized for playing but losing to good teams.

It also means that teams are not overly penalized for bad losses. In this way, the ranking is more focused on how good your team can be, not how bad your team can be.

I feel that come tournament time, when teams are focused and playing at their highest level, this will be a better indicator for success than overly penalizing them for a couple of bad mid-season losses when not playing at their full potential.

Anyway, let me know what you think!

Call It
01-25-2015, 06:40 PM
More fun with Excel spreadsheets!

I created my own computer ranking last night, called the GRaNT Computer Rankings -- (Stands for "Grant's Reasonable and Not Terrible Computer Rankings"). Pretty exciting stuff. I will post a BCI link with an easy-to-read ranking in the next couple days, but here is a temporary link (https://onedrive.live.com/redir?resid=EFA0679B9331D733!132&authkey=!ABrTxkHrWkMrDK0&ithint=file%2cxlsx).

Here's the jist of the rankings:

-All teams are assigned an arbitrary rating. I used .500 but it doesn't matter what you use because the rankings will come out the same regardless of what you start with.

-Each team's rating is recalculated based on their winning percentage and their opponent's rating as follows:
----A win adds the opponent's rating
----A loss adds zero
----A tie add's 1/2 the opponent's rating
------The sum is divided by the number of game's played.

Example: Teams A, B, and C each play teams X Y and Z once each.

Team A beats Team X, loses to Team Y, and ties Team Z. With starting value of 0.500 for everyone, Team A's new rating is ( 0.500 + 0 + 0.250 ) / 3 = 0.25.

Team B wins all 3 games. Team B's new rating is ( 0.500 + 0.500 + 0.500 ) / 3 = 0.500

Team C wins against Teams X and Y but loses to Team Z. Team C's new rating is ( 0.500 + 0.500 + 0 ) / 3 = 0.333

-That is done for all teams, and the new rating assigned to each team.

Then, the system restarts again with all of the new ratings instead of 0.500. It's done over and over until the ratings converge.

These are the rankings after the games of 1/24. I think Minnesota ends up leapfrogging BC with Wisconsin's win over Clarkson; I'll post an update once all the games are done.


Rank Team Rating
1 Boston College 0.8834
2 Minnesota 0.8769
3 Wisconsin 0.7308
4 Quinnipiac 0.6364
5 Harvard 0.6045
6 Boston University 0.5902
7 St. Lawrence 0.5463
8 Clarkson 0.5440
9 Cornell 0.5398
10 Minnesota-Duluth 0.5285
11 Bemidji State 0.5237
12 Mercyhurst 0.4539
13 Ohio State 0.4536
14 North Dakota 0.4528
15 Dartmouth 0.3992
16 Penn State 0.3853
17 Princeton 0.3569
18 Northeastern 0.3192
19 Maine 0.3086
20 Syracuse 0.3053
21 Vermont 0.2967
22 Robert Morris 0.2661
23 RIT 0.2541
24 Yale 0.2451
25 Lindenwood 0.2318
26 Connecticut 0.2231
27 Brown 0.1959
28 New Hampshire 0.1939
29 Providence 0.1909
30 Rensselaer 0.1600
31 Colgate 0.1597
32 Union 0.1552
33 St. Cloud State 0.1539
34 Minnesota State 0.0579


The one thing that ends up happening here is that you aren't really penalized for losing to good teams. A loss to the team in last place and a loss to the team in first place both affect your rating the same -- you get 0 points divided out by the extra game.

However, you are rewarded big time for playing really good teams -- you get a big boost to your rating for a win and aren't really hurt that much for a loss.

Anyway, let me know what you think!

BC #1?! Seems like fuzzy math

ManBehindTheCurtain
01-25-2015, 06:41 PM
Excellent contribution.

Any ranking system based on a rational examination of results on the ice is welcome. As with any system the assumptions at the base of the system are the only arguing point but I personally find no fault with your assumptions here.

The building of this ranking system clearly has taken more than the 4 or 5 minutes per week that most of us put into our personal rankings. I fear that either your employer or your family has borne the brunt of time lost from the time they expected of you. From what I know of you and from comments you have made here I suspect in this case your employer is the larger loser. If that is so I counsel you to be careful. But looking at the overall project and the personal resources that you have brought to the project I suspect that you are also bringing more than enough to the employer's table to justify the pittance that they pay you, probably considerably less than you deserve.

Very nice work.

Be very sure to take good care of your family.

TonyTheTiger20
01-25-2015, 06:51 PM
WOW well thank you, that is easily the nicest thing someone has said to me here!

Fortunately I actually did this one on the weekend with no time lost at work... for once :D

TonyTheTiger20
01-25-2015, 07:55 PM
Here are the updated ratings for today -- also I changed the scaling so that "0" means average. Same exact formula just scaled differently.

BC and Minnesota are so close that Wisconsin's win today over Clarkson moved Minnesota back on top.

Rank Team Rating
1 Minnesota 49.98
2 Boston College 48.71
3 Wisconsin 37.08
4 Quinnipiac 24.53
5 Harvard 21.21
6 Boston University 20.26
7 St. Lawrence 15.25
8 Cornell 14.52
9 Minnesota-Duluth 13.92
10 Clarkson 13.27
11 Bemidji State 12.45
12 North Dakota 9.56
13 Mercyhurst 6.33
14 Ohio State 6.10
15 Dartmouth 0.70
16 Penn State -0.51
17 Princeton -3.32
18 Maine -6.44
19 Syracuse -8.23
20 Northeastern -8.33
21 Vermont -9.91
22 Robert Morris -12.43
23 RIT -13.59
24 Yale -14.54
25 Connecticut -15.79
26 Lindenwood -15.91
27 Brown -19.09
28 New Hampshire -19.50
29 Providence -20.37
30 Rensselaer -22.84
31 Colgate -22.99
32 Union -23.23
33 St. Cloud State -23.68
34 Minnesota State -33.18

hammar22
01-25-2015, 09:50 PM
WOW well thank you, that is easily the nicest thing someone has said to me here!

Fortunately I actually did this one on the weekend with no time lost at work... for once :D


Really cool :):)

ARM
01-25-2015, 11:23 PM
My guess is that this one performs better as the season progresses. Early, it would be at the mercy of the unbalanced schedule, such that late starters like the Ivy League teams would be more undervalued than usual.

Eeyore
01-26-2015, 01:46 AM
This is somewhat similar to the way that KRACH is calculated, especially in its approach of trying to derive a team's rating from its expected outcome in each individual game. Th main conceptual difference is that KRACH weights losses by the quality of the opponent as well as the wins, which I think is a strength of KRACH. On the other hand, the calculations are much simpler in this method so people have a better idea of what is actually happening in the process.

TonyTheTiger20
01-26-2015, 11:53 AM
My guess is that this one performs better as the season progresses. Early, it would be at the mercy of the unbalanced schedule, such that late starters like the Ivy League teams would be more undervalued than usual.
Of course -- what ranking wouldn't? :eek:

(Other than Rutter, who includes previous season results early in the season and gradually weans them out)

TonyTheTiger20
01-26-2015, 11:55 AM
This is somewhat similar to the way that KRACH is calculated, especially in its approach of trying to derive a team's rating from its expected outcome in each individual game. Th main conceptual difference is that KRACH weights losses by the quality of the opponent as well as the wins, which I think is a strength of KRACH. On the other hand, the calculations are much simpler in this method so people have a better idea of what is actually happening in the process.
I really wish there was some kind of explanation for how KRACH is calculated. I mean I know there are explanations for what the numbers mean and that it's an iterative process and whatnot, but I haven't been able to actually find the method used to carry out the process. Same with Rutter and WCHODR. Great rankings, I assume, but I wanna know the nuts and bolts!

That's exactly almost entirely why I was motivated to create my own.

Eeyore
01-26-2015, 12:40 PM
I really wish there was some kind of explanation for how KRACH is calculated.

Have fun. (http://www.elynah.com/tbrw/tbrw.cgi?krach) There's a reason I tend to talk in analogies when trying to describe what KRACH is doing.

knight7
01-26-2015, 01:00 PM
I really wish there was some kind of explanation for how KRACH is calculated. I mean I know there are explanations for what the numbers mean and that it's an iterative process and whatnot, but I haven't been able to actually find the method used to carry out the process. Same with Rutter and WCHODR. Great rankings, I assume, but I wanna know the nuts and bolts!

That's exactly almost entirely why I was motivated to create my own.

TTT,

I grabbed this for you to read on how KRACH works

KRACH is short for “Ken's Rating for American College Hockey.” Ken is Ken Butler, a statistician, and the mathematical model he used is known as the Bradley-Terry Rating System. The system and its details are well documented in great mathematical detail online (see KRACH explanation at Ken Butler's homepage).

http://www.mghca.com/page/show/427504-krach-rankings. Click on the KRACH were it says explanation at Ken Butler's homepage.

TonyTheTiger20
01-26-2015, 01:17 PM
Oh you guys are just the best.

I'm glad this was after I created my own system, maybe I wouldn't gave felt like I had to lol

TonyTheTiger20
01-28-2015, 08:47 AM
I officially published the new ranking on BCI, and created a men's hockey version as well -- here's the link with a better explanation of how it's calculated and why.

http://www.bcinterruption.com/boston-college-hockey/2015/1/28/7925221/introducing-the-mens-womens-hockey-grant-computer-rankings-ncaa-frozen-four

Also I removed the "multiply by 100" scaling, there wasn't really a reason for it.

pokechecker
01-28-2015, 01:34 PM
I find it interesting that you list the men first

also, you have so much data and so many ads, I think you are shooting yourself in the foot if your goal is to have as many people as possible access your site, many people don't have much patience

kind of reminds me of the myspace days when people would put so much crap in the background that the message they were trying to convey got lost

on the one hand, I appreciate any new ranking system because it makes mine more accurate (in theory) however, looking at your rankings makes me question how much validity it has

TonyTheTiger20
01-28-2015, 01:50 PM
I find it interesting that you list the men first
1) "Men" comes before "women" alphabetically
2) More people care about men's hockey


also, you have so much data and so many ads, I think you are shooting yourself in the foot if your goal is to have as many people as possible access your site, many people don't have much patience

kind of reminds me of the myspace days when people would put so much crap in the background that the message they were trying to convey got lostNot my site, I don't care, I just write for them. ;)

TonyTheTiger20
01-28-2015, 02:33 PM
looking at your rankings makes me question how much validity it has
Biggest difference from KRACH is mine has UMD much lower and BU much higher despite having very similar records -- KRACH has one 6th and one 9th and mine has them reversed.

I'm guessing the reason why is that BU's losses are to worse teams -- BU has losses to to Maine, Northeastern, and Vermont, while UMD's three worst losses are to NoDak, Bemidji, and Bemidji.

These rankings rate all losses equally with the end result being that beating good opponents is rewarded rather than losing to bad opponents is penalized.

BU has a sweep over Clarkson, UMD has a sweep over Cornell. That's effectively a wash.

BU's best win is against QU. Duluth can't match that. They do have wins against Bemidji and North Dakota, but those are somewhat mitigated by the fact they they've also lost to those same teams.

Anyway there are a ton of little things you can look at. I think it gives a better idea of what a team's full-throttle capability is. Knowing that BU lost to Maine is less valuable information come tournament time than knowing that they beat QU because teams tend not to overlook opponents and lay total stinkers in the tournament.

Anyway, that's why the ranking is what it is. You can like or dislike that premise but saying "I'm looking at it and I don't like it" makes you sound a bit uninformed and eager to needlessly criticize and/or troll :) The rest of your post obviously just drove that point home.

brookyone
01-28-2015, 02:51 PM
Anyway, that's why the ranking is what it is. You can like or dislike that premise but saying "I'm looking at it and I don't like it" makes you sound a bit uninformed and eager to needlessly criticize and/or troll :) The rest of your post obviously just drove that point home.
I'm not aiming this specifically at you, but at the populace in general -- Why is it that people think that their right to an opinion means they have a right not to have their opinion criticized?

TonyTheTiger20
01-28-2015, 03:00 PM
I'm not aiming this specifically at you, but at the populace in general -- Why is it that people think that their right to an opinion means they have a right not to have their opinion criticized?I'm skeptical. Maybe the difference is some consider the manner of criticism while others do not. I've been around the forum long enough to know that there is a subjective aspect to an opinions acceptance or refutation, and the manner of refutation in particular. Prominent contributors having variable standards of decorum based on affiliations.

brookyone
01-28-2015, 03:04 PM
I have to say I'm flattered that you thought so highly of that post that you saved it for future use :eek:
Didn't save it. You can search relatively easily.

It's shocking you would find a neat way to qualify all to your "objective" benefit.