HFBoards

Go Back   HFBoards > NHL Eastern Conference > Metropolitan Division > New York Rangers
Mobile Hockey's Future Become a Sponsor Site Rules Support Forum vBookie Page 2
Notices

The Advanced Stats Thread Episode IV: A New Hope For Advanced Stats

Reply
 
Thread Tools
Old
12-07-2016, 04:20 PM
  #1
Gardner McKay
Global Moderator
Rangers this year
 
Gardner McKay's Avatar
 
Join Date: Jun 2007
Location: SoutheastOfDisorder
Country: United States
Posts: 15,872
vCash: 50
The Advanced Stats Thread Episode IV: A New Hope For Advanced Stats

Continue.

Part III

__________________
--If we had no faults of our own, we should not take so much pleasure in noticing those in others. - Francois de La Rochefoucauld
--
Gardner McKay is offline   Reply With Quote
Old
12-07-2016, 04:29 PM
  #2
Irishguy42
Registered User
 
Irishguy42's Avatar
 
Join Date: Sep 2015
Location: Bayside
Country: United States
Posts: 8,457
vCash: 500
GOAT title

Irishguy42 is offline   Reply With Quote
Old
12-07-2016, 05:29 PM
  #3
chosen
Registered User
 
chosen's Avatar
 
Join Date: Aug 2005
Posts: 6,734
vCash: 500
"So two weeks of play by the Blue Jackets holds more weight than a decade of top 5 CF% teams winning the Cup?"

Are you saying that one of the top 5 in the regular season or the playoffs?

chosen is offline   Reply With Quote
Old
12-07-2016, 05:30 PM
  #4
Hunter Gathers
Whiskey breath
 
Hunter Gathers's Avatar
 
Join Date: Feb 2002
Location: NJ/NYC
Country: United States
Posts: 99,640
vCash: 500
Send a message via AIM to Hunter Gathers
"The names are corsi and fenwick literally because of the people that implemented their usage. Jim Corsi and Matt Fenwick. lol it's not that hard to understand."

Erm. No kidding. The point was that such names are meaningless. Name the stat after what it shows, tracks, etc. Giving it some random name or naming it after a finely mustachioed fellow is moronic.

Hunter Gathers is offline   Reply With Quote
Old
12-07-2016, 05:41 PM
  #5
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Quote:
Originally Posted by Hunter Gathers View Post
"The names are corsi and fenwick literally because of the people that implemented their usage. Jim Corsi and Matt Fenwick. lol it's not that hard to understand."

Erm. No kidding. The point was that such names are meaningless. Name the stat after what it shows, tracks, etc. Giving it some random name or naming it after a finely mustachioed fellow is moronic.
Do you also hate the watt?

eyjee is offline   Reply With Quote
Old
12-07-2016, 06:17 PM
  #6
an Girar i
Registered User
 
Join Date: Oct 2014
Posts: 2,839
vCash: 500
Quote:
Originally Posted by Hunter Gathers View Post
"The names are corsi and fenwick literally because of the people that implemented their usage. Jim Corsi and Matt Fenwick. lol it's not that hard to understand."

Erm. No kidding. The point was that such names are meaningless. Name the stat after what it shows, tracks, etc. Giving it some random name or naming it after a finely mustachioed fellow is moronic.


cant forget this beaut either


an Girar i is offline   Reply With Quote
Old
12-07-2016, 08:13 PM
  #7
Hunter Gathers
Whiskey breath
 
Hunter Gathers's Avatar
 
Join Date: Feb 2002
Location: NJ/NYC
Country: United States
Posts: 99,640
vCash: 500
Send a message via AIM to Hunter Gathers
Quote:
Originally Posted by eyjee View Post
Do you also hate the watt?

__________________
"Of course giving Sather cap space is like giving teenagers whiskey and car keys." - SBOB
"Watching Sather build a team is like watching a blind man with no fingers trying to put together an elaborate puzzle." - Shadowtron
My MS Excel crashed so I couldn't watch the game. - ReggieDunlop68
Hunter Gathers is offline   Reply With Quote
Old
12-07-2016, 08:22 PM
  #8
Mac n Gs
Gorton plz
 
Mac n Gs's Avatar
 
Join Date: Jan 2014
Country: United States
Posts: 10,107
vCash: 500
Quote:
Originally Posted by Hunter Gathers View Post
"The names are corsi and fenwick literally because of the people that implemented their usage. Jim Corsi and Matt Fenwick. lol it's not that hard to understand."

Erm. No kidding. The point was that such names are meaningless. Name the stat after what it shows, tracks, etc. Giving it some random name or naming it after a finely mustachioed fellow is moronic.
To be fair, that Glicko **** was something Manny posted and I saw us at the top so I posted it here.

Kudos to silverfish those for using SAT

Mac n Gs is offline   Reply With Quote
Old
12-07-2016, 08:31 PM
  #9
Trxjw
Retired.
 
Join Date: May 2007
Location: Land of no calls..
Country: United States
Posts: 20,543
vCash: 500
Quote:
Possession time has already been proven to be a less effective form of analysis to shot attempts. That's why most analysts aren't saying that shot attempts are a proxy for possession anymore. They're not calling them possession stats. They are their own, better, thing.
Thank god it only took a few years of saying they measure puck possession for them to change their minds.

Trxjw is offline   Reply With Quote
Old
12-07-2016, 09:05 PM
  #10
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Quote:
Originally Posted by Hunter Gathers View Post
Why hello, nightmares, it has been a while.


eyjee is offline   Reply With Quote
Old
12-07-2016, 09:58 PM
  #11
Mac n Gs
Gorton plz
 
Mac n Gs's Avatar
 
Join Date: Jan 2014
Country: United States
Posts: 10,107
vCash: 500
Really wish there were more available/tracked neutral zone defensive stats. IMO, it's the one of the biggest drivers of possession.

Mac n Gs is offline   Reply With Quote
Old
12-08-2016, 01:55 AM
  #12
Machinehead
Our Bayley is better
 
Machinehead's Avatar
 
Join Date: Jan 2011
Location: NYC
Country: United States
Posts: 63,580
vCash: 50
Quote:
Originally Posted by chosen View Post
"So two weeks of play by the Blue Jackets holds more weight than a decade of top 5 CF% teams winning the Cup?"

Are you saying that one of the top 5 in the regular season or the playoffs?
Regular season

__________________
Machinehead is online now   Reply With Quote
Old
12-08-2016, 08:59 PM
  #13
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Kreider sucks
Pirri sucks
Hank sucks

Meanwhile, the Rangers shot attempt metrics look like if a Japanese guy who doesn't understand conversion rates tried to play cliff hangers on the price is right.

eyjee is offline   Reply With Quote
Old
12-09-2016, 12:38 AM
  #14
Machinehead
Our Bayley is better
 
Machinehead's Avatar
 
Join Date: Jan 2011
Location: NYC
Country: United States
Posts: 63,580
vCash: 50
Girardi's quality of teammates is 3.3 points higher than anyone else on the team

AV is basically doing this to him



And he's still.....Girardi.

Machinehead is online now   Reply With Quote
Old
12-09-2016, 10:20 AM
  #15
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Somebody needs to hire Manny already. This is just cutting edge work...

http://www.corsica.hockey/blog/2016/...ce-algorithms/

eyjee is offline   Reply With Quote
Old
12-09-2016, 10:49 AM
  #16
Adam Clendening
#RIPClendening
 
Adam Clendening's Avatar
 
Join Date: Apr 2015
Location: Press Box w/ Clendo
Country: United States
Posts: 9,004
vCash: 500
Quote:
Originally Posted by eyjee View Post
Somebody needs to hire Manny already. This is just cutting edge work...

http://www.corsica.hockey/blog/2016/...ce-algorithms/
Well don't bank on the Rangers doing it.

Adam Clendening is online now   Reply With Quote
Old
12-09-2016, 10:50 AM
  #17
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Quote:
Originally Posted by Adam Clendening View Post
Well don't bank on the Rangers doing it.
My real dream is for the Rangers to hire me so I can hire Manny and then fire myself

eyjee is offline   Reply With Quote
Old
12-09-2016, 11:14 AM
  #18
Mac n Gs
Gorton plz
 
Mac n Gs's Avatar
 
Join Date: Jan 2014
Country: United States
Posts: 10,107
vCash: 500
Quote:
Originally Posted by eyjee View Post
Somebody needs to hire Manny already. This is just cutting edge work...

http://www.corsica.hockey/blog/2016/...ce-algorithms/
I wonder if Manny isn't getting hired because of his outspoken views regarding other topics, which quite frankly, is a load of a horse crap if that's what's holding him back from getting an NHL job. Guy's a really, really good mathematician/statistician.

Mac n Gs is offline   Reply With Quote
Old
12-09-2016, 03:10 PM
  #19
Greg02
Registered User
 
Greg02's Avatar
 
Join Date: Jun 2009
Posts: 2,122
vCash: 500
I need to read it carefully, but 390-dimensional space is a huge red flag to me because of curse of dimensionality. Intuitively I'd be pretty surprised if there's something legitimately meaningful there.

Greg02 is offline   Reply With Quote
Old
12-09-2016, 04:17 PM
  #20
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500

eyjee is offline   Reply With Quote
Old
12-09-2016, 04:24 PM
  #21
Mac n Gs
Gorton plz
 
Mac n Gs's Avatar
 
Join Date: Jan 2014
Country: United States
Posts: 10,107
vCash: 500
Someone get Stathletes to show him how well G is playing and that Tanev is poop. Girardi for Tanev E5

Mac n Gs is offline   Reply With Quote
Old
12-09-2016, 04:26 PM
  #22
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Quote:
Originally Posted by Mac n Gs View Post
Someone get Stathletes to show him how well G is playing and that Tanev is poop. Girardi for Tanev E5
seriously!

I can't imagine Stathletes image is that tarnished, though, considering the Chayka hiring and that they are still very much so in business.

eyjee is offline   Reply With Quote
Old
12-09-2016, 04:52 PM
  #23
Mac n Gs
Gorton plz
 
Mac n Gs's Avatar
 
Join Date: Jan 2014
Country: United States
Posts: 10,107
vCash: 500
Quote:
Originally Posted by eyjee View Post
seriously!

I can't imagine Stathletes image is that tarnished, though, considering the Chayka hiring and that they are still very much so in business.
Wait that's Chayka's company? LOL

Mac n Gs is offline   Reply With Quote
Old
12-09-2016, 07:25 PM
  #24
Greg02
Registered User
 
Greg02's Avatar
 
Join Date: Jun 2009
Posts: 2,122
vCash: 500
Gave a real read of the PSO article. Don't have hard evidence, but based on what was shared it's worthless.

Disclaimer: ML isn't really my area of CS, and stats really isn't my area of math.

That said, one of the reasons I dislike the ML field is because of how much buzzwordy it is. People go "oh, ML is the cool new thing, so let's just blindly use some standard library to solve all our problems" and then go and come up with things like this. It's not that I don't think ML can be applied in effective ways to hockey, because I do. It's just that it requires some careful thought.

For instance, if you're throwing in four hundred dimensions into your model, it is almost certain that some of them are going to be correlated (and in fact, I guarantee you that a number of these dimensions are pairwise correlated in extremely trivial ways). Now you're solving an optimization problem where the solution space isn't unique. So given different initial locations and velocities, you'll likely get different weights. Notice that calculating the weights was only ran once. But because the dimensions are linearly dependent, there are multiple "correct" convergences. There are a few interesting pieces of data that aren't included in the blog post; one might be to show the actual weights of each dimension. Then run it multiple times and compare. The multi-colinearity can be addressed by PCA or something, but honestly, better off with aggressive feature selection. So many features just seems rife for overfitting to me.

The model is then evaluated on its training data, which is a worthless judgment. Obviously it's going to converge so that the training set is fulfilled; it took those exact pairwise comparisons as the reason to weigh the different features in the first place! It's circular. The test dataset has to be different from the training dataset.

Greg02 is offline   Reply With Quote
Old
12-09-2016, 07:47 PM
  #25
eyjee
the basement
 
eyjee's Avatar
 
Join Date: Jun 2008
Location: Alphabetrium
Country: United States
Posts: 27,946
vCash: 500
Quote:
Originally Posted by Greg02 View Post
Gave a real read of the PSO article. Don't have hard evidence, but based on what was shared it's worthless.

Disclaimer: ML isn't really my area of CS, and stats really isn't my area of math.

That said, one of the reasons I dislike the ML field is because of how much buzzwordy it is. People go "oh, ML is the cool new thing, so let's just blindly use some standard library to solve all our problems" and then go and come up with things like this. It's not that I don't think ML can be applied in effective ways to hockey, because I do. It's just that it requires some careful thought.

For instance, if you're throwing in four hundred dimensions into your model, it is almost certain that some of them are going to be correlated (and in fact, I guarantee you that a number of these dimensions are pairwise correlated in extremely trivial ways). Now you're solving an optimization problem where the solution space isn't unique. So given different initial locations and velocities, you'll likely get different weights. Notice that calculating the weights was only ran once. But because the dimensions are linearly dependent, there are multiple "correct" convergences. There are a few interesting pieces of data that aren't included in the blog post; one might be to show the actual weights of each dimension. Then run it multiple times and compare. The multi-colinearity can be addressed by PCA or something, but honestly, better off with aggressive feature selection. So many features just seems rife for overfitting to me.

The model is then evaluated on its training data, which is a worthless judgment. Obviously it's going to converge so that the training set is fulfilled; it took those exact pairwise comparisons as the reason to weigh the different features in the first place! It's circular. The test dataset has to be different from the training dataset.
This is interesting and always appreciate your insights.

One thing that I can comment on is overall Machine Learning. I read somewhere recently that most organizations think they want ML, but really all they need is cleaned up data and a linear regression. I believe that Machine Learning is going to start coming to my industry (web analytics) very, very soon, and I think it'll be overkill.

I've been trying my hardest to get a couple of our departments within the agency to allow me to run their data viz or linear modeling, as I think that's the next logical step for us.

Found it: http://www.overcomingbias.com/2016/1...also-bust.html



Last edited by eyjee: 12-09-2016 at 08:11 PM.
eyjee is offline   Reply With Quote
Reply

Forum Jump


Bookmarks

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off



All times are GMT -5. The time now is 12:45 AM.

monitoring_string = "e4251c93e2ba248d29da988d93bf5144"

vBulletin Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.
HFBoards.com is a property of CraveOnline Media, LLC, an Evolve Media, LLC company. 2017 All Rights Reserved.