Burkie's Challenge Accepted: Lets Make An Analytics System!

[Rutuu]

Originally Posted by Where ru Chris O'Sullivan

And in Baseball there is no physical interaction with other players. This is a huge part of the NHL. It's not only being able to skate, pass, shoot, but the ability to workaround others (the best in the world).

The systems aren't the problem...its the data that is collected. That's the opportunity though.

Goaltending and defenseman would be your best bets to concentrate on as the correlation between draft position and point production for forwards is pretty high, so you'd have to assume that most hockey people have it figured out.

Top goalies come from a lot of areas in the draft. Top defenseman come from the top 4 rounds...it seems like there is more opportunity to be better within the drafts for those two positions.

Goalies you need to figure out what drives sv% (since the difference between a 0.910 goalie and a 0.930 goalie is a playoff birth and a shot at Lord Stanley)...positioning, quickness (reaction time/athleticism) and puck direction (rebounds)?

1. Static positioning...where does the goalie position himself before the shot (Compare to an ideal model)

2. Dynamic positioning...where does the goalie position himself after the shot (in relation to the rebound...giving himself a chance at the second save...plus does he choose the optimal save type)
Ex: When in the butterfly does he lift the proper leg first after
Once again compare it to the ideal model

3. Where does the rebound go?...absorbed...or to a spot where it can be hammered in?

4. Add in a quickness component as well that has to do with reaction time and ability to go post to post.

Now you just need the ability to condense video footage of every goalie you want to evaluate and do the physics to figure out shot angles in a bunch of different rinks.

Or you do it like Feaster did and acquire a bunch of goalies for cheap and hope one emerges around the 0.920% mark for a season or two.

[wretched34]

Originally Posted by CalgaryFan1988

And I'll make sure everyone does their job safely.

Who is in charge of payroll?

I'll handle payroll, as soon as there is money to pay.

Until then, I'll just Analyze stuff.

[Saqe]

You would always need to quantify qualitative data which makes it impossible to make a system solely behind your computer screen... someone who is a good judge of personality, work ethic, heart etc. needs to actually go and see the players and talk to them.

A guy who can actually evaluate people on top of analyzing statistics probably would be on to something.

[AltaGuy]

I've always wanted to measure players' "clutch" factors. Maybe, goals and primary assists per TOI in the the third period when trailing or tied. Then you subtract GA per TOI in same situations, then you account for QoC, and then you have something that means something.

Ok, go. Haha

[BurningSteel]

E = MC squared / Iginla + Kipper x Monahan = 2????

Somethings off so I guess I am out.

[cannon7]

I'm a Computer Scientist with a background in ML (machine learning). I've spent some time looking into the potential for leveraging ML to find patterns for exactly this purpose. Unfortunately hockey isn't as quantifiable as, say, baseball -- so the Moneyball approach is far more limited. But it is still possible, assuming you can get the data you need.

Here's the bird's eye view...

It all starts with an algorithm, or a mathematical equation, where the input is a player's pre-draft statistics, the league they are in, the team they play from, where they were born, when they were born, height/weight (overtime, preferrably) and so on. the output is an accurate probability for success in the NHL.

The way you build this algorithm for ML purposes is that you provide all the above inputs and their respective properties. For example, you wouldn't ust have Petergorough, ON as a property -- instead you'd have the number of players born there, the number of players who made the NHL who were from there, etc. You plug these variables in, provide some basic rules (logic) by which the ML can alter the algorithm (of which there should be millions/billions of combinations) and then run it.

Now the way ML will determine if it has improved the algorithm or not is to compare the input with the output. So it will crunch all the existing data you can get your hands on as to a player's pre-draft stats and post-draft NHL success. It will then compare the output of the algorithm with reality. Now it does this in aggregate, meaning it compares thousands of inputs with thousands of outputs, since comparing just one player to reality would only be anecdotal.

Eventualy (billions of calculations later) it will come to a point where it cannot improve the algorithm any more given the dataset provided. You then take this algorithm and apply it to the current draft crop and see what you get.

[Enoch Root]

Start with two groups.

The first group (the people from this thread possibly?) are charged with developing and improving advanced stats for the purpose of improving analytics, with the ultimate goal of using said analytics in order to try and build a better team.

The second group (let's say Burke and Treliving for fun) are charged with building a better team, with the ultimate goal of ending up with better advanced stats for their players.

Two questions:

1) which group would you rather be in?

2) which group is more likely to realize the ultimate goal?

In other words, will improving your stats result in a better team? Or will improving your team result in better stats?

Does improving possession time result in more winning? Or does more winning have the side effect of more possession time?

[cannon7]

Just to be clear, the description above is a simplification of how ML really works. I don't nkow how many CS/SE types there are around here so I kept it simple.

I guess thepoint is that while such a system would be difficult to build given the scarcity of data -- specifically junior-level and prior. The reality is that you'd only need to achieve a measurable improvement over existing systems to be of value to scouts. Even if your system was only 5% more accurate in predicting a player's success in the NHL, you'd still be able to make a boatload of money from scouts looking for an edge.

The real data you'd want, ethical implications aside, is genealogical. If I could compare the genetic makeup of a player against thousands of NHLers I imagine I'd get some striking results. I think few would argue that there's a genetic component to a player's success in any sport.

[Enoch Root]

Originally Posted by cannon7

Just to be clear, the description above is a simplification of how ML really works. I don't nkow how many CS/SE types there are around here so I kept it simple.

I guess thepoint is that while such a system would be difficult to build given the scarcity of data -- specifically junior-level and prior. The reality is that you'd only need to achieve a measurable improvement over existing systems to be of value to scouts. Even if your system was only 5% more accurate in predicting a player's success in the NHL, you'd still be able to make a boatload of money from scouts looking for an edge.

The real data you'd want, ethical implications aside, is genealogical. If I could compare the genetic makeup of a player against thousands of NHLers I imagine I'd get some striking results. I think few would argue that there's a genetic component to a player's success in any sport.

Here's the rub with analytics though - even if you could create something that actually improved results by 5% (and I am not suggesting that you couldn't), if it actually did offer an advantage, all clubs would immediately employ it (or pursue similar sources), thereby eliminating the advantage that was created.

[getbak]

Originally Posted by Enoch Root

Here's the rub with analytics though - even if you could create something that actually improved results by 5% (and I am not suggesting that you couldn't), if it actually did offer an advantage, all clubs would immediately employ it (or pursue similar sources), thereby eliminating the advantage that was created.

That's exactly why Burke said he would pay cash to ensure you didn't sell it to any other team.

[cannon7]

Originally Posted by Enoch Root

Here's the rub with analytics though - even if you could create something that actually improved results by 5% (and I am not suggesting that you couldn't), if it actually did offer an advantage, all clubs would immediately employ it (or pursue similar sources), thereby eliminating the advantage that was created.

That's a fair point, but atleast as of today you can patent software. So if you were smart, if you somehow were able to come up with a method to improve existing scouting analytics you'd want to first patent it and then license it to the highest bidders. The arrangement being that Team A pays a premium so that you won't just go and license it to Teams B-Z. That's kind of how it is done now, there are companies that specialize in this and they have a product for everyone and then a product with the "secret sauce" only for those willing to pay top dollar for it.

[Enoch Root]

Originally Posted by getbak

That's exactly why Burke said he would pay cash to ensure you didn't sell it to any other team.

Yes I know. But if it can be created, it can be replicated.

[cannon7]

You can also patent algorithms. Say your ML system came up with five algorithms that outperformed existing scouting analytics. You could just patent all five of them, thus preventing someone else from going to market with the second best algorithm and under-cutting you.

That is, of course, assuming you could actually understand the algorithm you got back from ML. That's actually the hard part when it comes to patents. If you can't explain how it works then you can't patent it.

Human beings are limited by concepts, but a machine is only concerned with numbers, so understanding how ML drew a certain conclusion void of the concepts behind them is difficult.

Assuming you can't explain how the algorithm works, your best bet is tolock it up as best you can and hope no one smarter than you stumbles upon it and is able to explain it in patentable form.

Unfortunately there's always someone smarter than you out there.

[schooner]

Originally Posted by MattyC

Economics degree here with lots of stats courses and sports econ. If someone else wants to gather a bunch of data im in for helping with regressions and such.

Economics degree here as well, but I was drunk, stoned or asleep for most the 4 years (alright to be honest 5.5 years, buts who's counting?) so I might have limited use.


On topic though, sure you can build a standard model like you would for any statistical problem, the biggest challenge is the quality of the data. And that is hard to work with in hockey. The integrity of the data is key to coming up with a robust correlation. Too many different leagues, coaches, rules, measures of success, statisticians recording data etc, to come up with a reliable predictor of success.

[Iceman90]

Keep up the good work team!

[#-3]

It there a good place to find more in depth stats on lower leagues, it would be good it if could be sorted, by league, by season.

Like:

TOI
PPG
F/O taken
F/O Pct
Shot Attpemts
Shots
Hits

And the Normal ones: GP, G, A, P, +/-, PIM

I think I could reverse engineer a formula from past stats, it would hard to disprove. By checking it historically it would be correct, and it would take years to prove it wrong.

[simmer2]

What does the NFL have for a data analytics system? I would think they have some pretty superior data software systems.

[GGG]

The answer is the sports VU cameras installed in every NBA areana. They track everying. Speed, acceleration, positioning. There is so much data it is out of the realm of the hobbyist. If these cams were put in the NHL you could determine shot quality by scoring location, velocity, screening. It is the first advanced, advanced stats as it stops relying on humans counting things as proxys for other things. All of the real data would be there.

Ideally you take that dataset and search for corrolations that produce positive outcomes. You develop a score for each variable so each unique ice position has a probability of scoring. Players that on average increase this value should be better than players who dont. Put these cameras in a few jr rinks too and you can find what things drive nhl success in 10 years. Do this first and you have an edge. Any team that shares an arena with an NBA already has the cameras, its just making them work for hockey.

[SebC]

Originally Posted by Enoch Root

The second group (let's say Burke and Treliving for fun) are charged with building a better team, with the ultimate goal of ending up with better advanced stats for their players.

Burke and Treliving aren't tasked with improving the Flames advanced stats - their task is to improve the "bottom line" stats.

[Freeway]

Originally Posted by SebC

Burke and Treliving aren't tasked with improving the Flames advanced stats - their task is to improve the "bottom line" stats.

Wins.