The Advanced Stats Thread Episode V: Rick Nash Camera Stares/60

[silverfish]

Holy **** my model ****ing sucked for the 16-17 season.

Out of 298 evaluated forwards it predicted the correct amount of goals for 51 of those players, the correct amount of primary points for 66 players, and both goals and primary points for 25 players.

r^2 to goals: 0.1982
r^2 to pri points: 0.2952

Compare to what I had the model working last year of:
r^2 to goals: .3037
r^2 to pri points: .3421

****.

 

[Irishguy42]

Holy **** my model ****ing sucked for the 16-17 season.

Out of 298 evaluated forwards it predicted the correct amount of goals for 51 of those players, the correct amount of primary points for 66 players, and both goals and primary points for 25 players.

r^2 to goals: 0.1982
r^2 to pri points: 0.2952

Compare to what I had the model working last year of:
r^2 to goals: .3037
r^2 to pri points: .3421

****.

I felt like this past season was an odd one in general. Will be interesting to see how next year goes to compare it to this season.

 

[silverfish]

I felt like this past season was an odd one in general. Will be interesting to see how next year goes to compare it to this season.

I am so disappointed. It, for just some dude in a cubicle, did pretty well for the 15-16 season. Totally ****s the bed this year. So discouraging

(blue line is line of best fit, dashed line is x=y)

 

[SnowblindNYR]

Holy **** my model ****ing sucked for the 16-17 season.

Out of 298 evaluated forwards it predicted the correct amount of goals for 51 of those players, the correct amount of primary points for 66 players, and both goals and primary points for 25 players.

r^2 to goals: 0.1982
r^2 to pri points: 0.2952

Compare to what I had the model working last year of:
r^2 to goals: .3037
r^2 to pri points: .3421

****.

Is it possible that you added too many variables and had a problem with overfitting and you got a lot of noise?

 

[silverfish]

Is it possible that you added too many variables and had a problem with overfitting and you got a lot of noise?

It's possible, but everything is weighted. Inputs are weighted, and the player comparables are all weighted... It may be too lenient of a model, if anything, which I supposes could be overfitting.

What then tends to happen is the model predicts very close to league average, because it's taking the data from too many players, perhaps.

The lowest 82 game goal total it predicted was 3.75, Ryan Reaves.
The highest 82 game goal total it predicted was 14.1, Taylor Hall.

11 goals isn't enough of a difference between the league's best & worst 5v5 scorers [the difference in primary points was 20, league best to worst].

 

[Mac n Gs]

Wow, literally RIP in peace computer boys AND silverfish. Down with corsi, up with grit! "Advanced" stats, sad!

Don't get down silverfish, you'll tweak the model and crush it next year

 

[SnowblindNYR]

Wow, literally RIP in peace computer boys AND silverfish. Down with corsi, up with grit! "Advanced" stats, sad!

Don't get down silverfish, you'll tweak the model and crush it next year

His bad model>>>>>>>70% of hockey fans that think Tanner Glass is a good useful player

 

[Thirty One]

Holy **** my model ****ing sucked for the 16-17 season.

Out of 298 evaluated forwards it predicted the correct amount of goals for 51 of those players, the correct amount of primary points for 66 players, and both goals and primary points for 25 players.

r^2 to goals: 0.1982
r^2 to pri points: 0.2952

Compare to what I had the model working last year of:
r^2 to goals: .3037
r^2 to pri points: .3421

****.

..... I used to believe in things...

 

[Mac n Gs]

His bad model>>>>>>>70% of hockey fans that think Tanner Glass is a good useful player

Lord help us if there's > 70% of hockey fans that think Glass is a good player. Great teammate =/= good hockey player.

 

[SnowblindNYR]

Lord help us if there's > 70% of hockey fans that think Glass is a good player. Great teammate =/= good hockey player.

Probably not 70% BUT there are still a lo of meathead Rangers fans that still think it's the 90s and think that tough guys with no skill are great. I remember my Facebook lighting up with people complaining that Mike ****ing Rupp got traded.

 

[SnowblindNYR]

Also, I meant that his bad model is better than most hockey fans' knowledge of the game, and the greater than signs were meant to show that, it was meant as 70%, not >70%.

 

[SnowblindNYR]

I keep forgetting to post this. But for the stat nerds there's a good book that came out a few months back written by Michael Lewis of Moneyball, The Big Short, and The Blind Side fame. It's about two psychologists that did research on human judgment and decision making and basically came to the conclusion that people make a lot of systematic mistakes. They make mental shortcuts that don't always correspond with reality. The book is called "The Undoing Project".

The Undoing Project

Stats Nerds 1- Eyeball test Proponents 0

 

[Ola]

Thanks Snow, I can't recommend it since I haven't read it but I've read his first and big short and I can certainly recommend anything written by Lewis. He is a very good writer/storyteller, it's hard in a movie like Big Short, but he is good at explaining things basically.

 

[Ola]

People are proving that SVA CF% is a very good predictive stat, that's it.

Sorry for sounding a bit condecending, my bad, my point is just -- it's not going to rank extraordinary well compared to goals for, goals against, goal diff, wins, teams with good players and like any random -- but yet good indicator of a "good" team. Right?

Just saying, to come out very hard claiming that it's extraordinary that in a Conference Final and a Stanley Cup Final the team that takes more shots than it give up is going to win, is IMO just focusing a bit on the wrong things. Of course it's a good basis to predict future success on. Like many other factors (wins, goal diff and so forth). I just don't see the overwhelming result that arguably has a trickle down effect.

 

[Ola]

I keep forgetting to post this. But for the stat nerds there's a good book that came out a few months back written by Michael Lewis of Moneyball, The Big Short, and The Blind Side fame. It's about two psychologists that did research on human judgment and decision making and basically came to the conclusion that people make a lot of systematic mistakes. They make mental shortcuts that don't always correspond with reality. The book is called "The Undoing Project".

The Undoing Project

Stats Nerds 1- Eyeball test Proponents 0

It's interesting for sure this subject, I remember hearing on some Law Pod, Harvard Law probably, of how a reference group of top lawyers (Supreme Court judges, law makers, top professors etc) really badly got its arsed spanked by a big data formula when it comes to predicting the out come of a number of randomly selected cases in US courts. Think the score was like 75% vs 96% or something like that.

I think these are the type of mistakes people do when putting like Corsi on a pedestal. Hand made formula logically should make sense. Hallelujah.

I just believe that you aren't getting the really good stuff until you throw every single data you can get your hand on, from nationality and size and what not to goals, shot attempts and so forth, into a massively large data base and cross run it against actual success.

And even then, you have a significant problem with the game being ever changing and teams playing different styles. What's best for LAK isn't best for Chi and vice versa. Preferbly these styles should -- definitely -- be identified and added to the data base. The impact of system is too large to disregard from.

Stats of today are of course extremely useful in many areas. But lets be honest, they aren't quite there yet in hockey.

 

[SnowblindNYR]

Thanks Snow, I can't recommend it since I haven't read it but I've read his first and big short and I can certainly recommend anything written by Lewis. He is a very good writer/storyteller, it's hard in a movie like Big Short, but he is good at explaining things basically.

I read exclusively non-fiction and I think he's my favorite author. I've heard people criticize some of his work, most notably Flash Boys.

I read: The Big Short, Moneyball, The Blind Side, Flash Boys, and Liar's Poker.

I might be forgetting one here or there. My least favorite of his is Flash Boys and I've heard a number of people criticize it for being not very accurate.

 

[Ola]

Snowblind- Actually wasn't aware he wrote Moneyball! I listened to Liars Pokers as an audio book (haha love his voice) and read big short.

It strikes me that that is a heck of a collection of books. Darn. Basically up there with Cormac. And in a sense David Simon. Think I am going to start doing some needlestick lobbying for him to get the Nobel Prize. Would be really cool.

And to be honest, I don't really care about accuracy. Are Liars Pokers and Big Short "accurate" descriptions of Wall Street. Of course not in many senses. But they really help raising the understanding of Wall Street and how it functions in some situations. It the same thing but still very important.

 

[SnowblindNYR]

It's interesting for sure this subject, I remember hearing on some Law Pod, Harvard Law probably, of how a reference group of top lawyers (Supreme Court judges, law makers, top professors etc) really badly got its arsed spanked by a big data formula when it comes to predicting the out come of a number of randomly selected cases in US courts. Think the score was like 75% vs 96% or something like that.

I think these are the type of mistakes people do when putting like Corsi on a pedestal. Hand made formula logically should make sense. Hallelujah.

I just believe that you aren't getting the really good stuff until you throw every single data you can get your hand on, from nationality and size and what not to goals, shot attempts and so forth, into a massively large data base and cross run it against actual success.

And even then, you have a significant problem with the game being ever changing and teams playing different styles. What's best for LAK isn't best for Chi and vice versa. Preferbly these styles should -- definitely -- be identified and added to the data base. The impact of system is too large to disregard from.

Stats of today are of course extremely useful in many areas. But lets be honest, they aren't quite there yet in hockey.

That may be true but I believe studies have shown that Corsi correlates well with winning a cup. Don't quote me on that. I don't think people just blindly follow Corsi because it sounds good. Also, sometimes having too many variables can hurt a model as it could start picking up noise.

 

[SnowblindNYR]

Snowblind- Actually wasn't aware he wrote Moneyball! I listened to Liars Pokers as an audio book (haha love his voice) and read big short.

It strikes me that that is a heck of a collection of books. Darn. Basically up there with Cormac. And in a sense David Simon. Think I am going to start doing some needlestick lobbying for him to get the Nobel Prize. Would be really cool.

And to be honest, I don't really care about accuracy. Are Liars Pokers and Big Short "accurate" descriptions of Wall Street. Of course not in many senses. But they really help raising the understanding of Wall Street and how it functions in some situations. It the same thing but still very important.

Well Liar's Poker is sort of a comedy book, I think. So I don't care if that's exaggerated. I don't like the Big Short being exaggerated though. I wonder what kind of impact on policy a book like that has. If it does have any impact, I want it to be as close to truthful as possible. To his credit, I read another book on the topic (that one focuses on John Paulson the biggest player in the whole thing in addition to Michael Burry and the rest), and it read similarly.

I really like the way his books read. They're written in a very conversational way that makes them an easy read.

 

[ManUtdTobbe]

Sorry for sounding a bit condecending, my bad, my point is just -- it's not going to rank extraordinary well compared to goals for, goals against, goal diff, wins, teams with good players and like any random -- but yet good indicator of a "good" team. Right?

Just saying, to come out very hard claiming that it's extraordinary that in a Conference Final and a Stanley Cup Final the team that takes more shots than it give up is going to win, is IMO just focusing a bit on the wrong things. Of course it's a good basis to predict future success on. Like many other factors (wins, goal diff and so forth). I just don't see the overwhelming result that arguably has a trickle down effect.

Depends, SVA CF% will predict what will happen MUCH better then GF/GA/Goal diff etc... It's a good predictive stat, not a good descriptive stat, this is what "most people" doesn't seem to be able to differentiate between. Wins, goal diff etc are much worse at predicting future success.

 

[Ail]

Probably not 70% BUT there are still a lo of meathead Rangers fans that still think it's the 90s and think that tough guys with no skill are great. I remember my Facebook lighting up with people complaining that Mike ****ing Rupp got traded.

Those nubmers are skewed. Dagoon has a lot of alternate accounts.

 

[Ola]

Depends, SVA CF% will predict what will happen MUCH better then GF/GA/Goal diff etc... It's a good predictive stat, not a good descriptive stat, this is what "most people" doesn't seem to be able to differentiate between. Wins, goal diff etc are much worse at predicting future success.

Is it really?

Like that table show that the avg adjusted CF% of the top 4 teams in the POs, after Christmas, is what 8.5 overall in the NHL (7.25 overall the last 4 years). What is the avg. goal diff of the top 4 team each year? A quick check and I get the following results:

2016:
Pitt 4th
TB 5th
SJ 6th
STL 8th

2015:
Chi 8th
TB 2nd
NYR 1st
Ana 13th

2014
LAK 4th
NYR 9th
Chi 3rd
MTL 17th

2013
Chi 1st
Bos 5th
Pitt 2nd
Ana 4th

Avg. 5.75 overall. Looking at goal differential during the regular season seem to be a much better indicator of future success than Corsi. Or? What am I missing? I am not saying that Corsi isn't an indicator of future success, but every once in a while you see a tweet claiming how Corsi is by far the best, but those tweets always at least to me give the impression of being biased to be honest. First stat I pick beats Corsi by a mile. Why is that? Did nobody think of goal differential? Its a pretty obvious stat, displayed right there in the standings. How about GF? Wins? GA? Depth scoring?

 

[silverfish]

The Undoing Project is a great read. Would recommend.

 

[Ola]

The Undoing Project is a great read. Would recommend.

Yeah I got to pick it up along with Flash Boys.

If was a GM of an NHL team, I would try to hire someone writing the algoritms of the HFT robots. Just dump as much Data as you can into a sample size, then cross run it against de facto Results. Then you can speak about the numbers don't lying, if you have an enormous amount of data. With these hand made formulas, you might get the "brain" of big data but the "eyes" and the "neck" deciding which direction the eyes are looking are still just as influenced by eye test and subjective opinions as any other basis for an opinion.

 

[ManUtdTobbe]

Is it really?

Like that table show that the avg adjusted CF% of the top 4 teams in the POs, after Christmas, is what 8.5 overall in the NHL (7.25 overall the last 4 years). What is the avg. goal diff of the top 4 team each year? A quick check and I get the following results:
..... Wall of numbers

I'll leave this open for someone else to answer if they feel like it since i'm not good enough at digging up numbers and explaining how/why etc.