News & Notes XLV: Simply Having A Wonderful Canesmastime

[Discipline Daddy]

"Hey, nerdy hockey kid. Want your name on the Stanley Cup?"

Candidate A: "I think you need more size in front on your power play."

Candidate B: "I think you need to find a better PPQB."

Candidate C: "Special teams (i.e. power play and penalty kill) situations play an outsized role in determining the outcome of ice hockey games. Yet, quantitative methods for characterizing special teams tactics are limited. This work focuses on team structure and player deployment during in-zone special teams possessions. Leveraging player and puck tracking data from the National Hockey League (NHL), a framework is developed for describing player positioning during 5-on-4 power play and 4-on-5 penalty kill possessions. More specifically, player roles are defined directly from the player tracking data using non-negative matrix factorization, and every player is allocated a unique role at every frame of tracking data by solving a linear assignment problem. Team formations naturally arise through the combination of roles occupied in a frame. Roles that vary on a per-frame basis allow for a fine-grained analysis of team structure. This property of the roles-based representation is used to group together similar power play possessions using latent Dirichlet allocation, a topic modelling technique. The concept of assignments, which remain constant over an entire possession, is also introduced. Assignments provide a more stable measure of player positioning, which may be preferable when assessing deployment over longer periods of time."

latent Dirichlet allocation? It's legit, but you're bordering on Turbo Encabulator territory here.

 

[Nikishin al Gaib]

....and then screws us with a 6pm game in the middle of rush hour traffic during a weekday for Game 2

Sometimes I really do think this league sets us up for failure/criticism. Easter Sunday afternoon game and a game facing off during rush hour in a southern Bible Belt city without mass public transportation is horseshit when you think about it.

 

[AD Skinner]

"Hey, nerdy hockey kid. Want your name on the Stanley Cup?"

Candidate A: "I think you need more size in front on your power play."

Candidate B: "I think you need to find a better PPQB."

Candidate C: "Special teams (i.e. power play and penalty kill) situations play an outsized role in determining the outcome of ice hockey games. Yet, quantitative methods for characterizing special teams tactics are limited. This work focuses on team structure and player deployment during in-zone special teams possessions. Leveraging player and puck tracking data from the National Hockey League (NHL), a framework is developed for describing player positioning during 5-on-4 power play and 4-on-5 penalty kill possessions. More specifically, player roles are defined directly from the player tracking data using non-negative matrix factorization, and every player is allocated a unique role at every frame of tracking data by solving a linear assignment problem. Team formations naturally arise through the combination of roles occupied in a frame. Roles that vary on a per-frame basis allow for a fine-grained analysis of team structure. This property of the roles-based representation is used to group together similar power play possessions using latent Dirichlet allocation, a topic modelling technique. The concept of assignments, which remain constant over an entire possession, is also introduced. Assignments provide a more stable measure of player positioning, which may be preferable when assessing deployment over longer periods of time."

I had a professor in grad school who liked to say he graded by how much the paper ended up weighing

 

[Navin R Slavin]

latent Dirichlet allocation? It's legit, but you're bordering on Turbo Encabulator territory here.

Look, bro, if you're not keeping up with the latest generative statistical models in order to keep up your end of the hockey board conversations, that's on you.

(PS: here's his paper: Spatial roles in hockey special teams)

(PPS: what this abstract implies is that the Canes might actually be building an LLM-style recommender system to answer critical hockey questions -- an AI Coach, basically. Which is *incredibly* audacious, and if successful, basically turns the "coach" into the guy that can most successfully translate the AI coach's analysis into actual player mechanics.)

 

[Stanky and Janky]

....and then screws us with a 6pm game in the middle of rush hour traffic during a weekday for Game 2

Sometimes I really do think this league sets us up for failure/criticism. Easter Sunday afternoon game and a game facing off during rush hour in a southern Bible Belt city without mass public transportation is horseshit when you think about it.

Didn't seem to hurt the attendance or atmosphere on Sunday. And we're far from the only team getting some questionable start times this postseason in the name of the almighty television.

 

[tarheelhockey]

....and then screws us with a 6pm game in the middle of rush hour traffic during a weekday for Game 2

Sometimes I really do think this league sets us up for failure/criticism. Easter Sunday afternoon game and a game facing off during rush hour in a southern Bible Belt city without mass public transportation is horseshit when you think about it.

All part of being a small market.

The alternative is getting the Leafs/Bruins/Habs/Rangers fanbase experience, which is absolutely not a trade I’d make.

 

[Bunch of Jurcos]

Look, bro, if you're not keeping up with the latest generative statistical models in order to keep up your end of the hockey board conversations, that's on you.

(PS: here's his paper: Spatial roles in hockey special teams)

(PPS: what this abstract implies is that the Canes might actually be building an LLM-style recommender system to answer critical hockey questions -- an AI Coach, basically. Which is *incredibly* audacious, and if successful, basically turns the "coach" into the guy that can most successfully translate the AI coach's analysis into actual player mechanics.)

I get that you were being tongue and cheek here but I'm actually surprised that we're not using more real time stats to make coaching decisions. I believe that statistics would make for a more interesting game, provided the players listen, there would be more scoring opportunities, more breakaways due to noticing tendencies, and higher scoring games due to better information on how a goalie is stopping shots. Is this type of information even legal to use during a game?

All of that is predicated on the fact that the coach isn't too stubborn or prideful to actually make use of the data.

 

[Unhinged Finn]

basically turns the "coach" into the guy that can most successfully translate the AI coach's analysis into actual player mechanics.)

AI coach: bench Burns
RBA:

 

[WreckingCrew]

I get that you were being tongue and cheek here but I'm actually surprised that we're not using more real time stats to make coaching decisions. I believe that statistics would make for a more interesting game, provided the players listen, there would be more scoring opportunities, more breakaways due to noticing tendencies, and higher scoring games due to better information on how a goalie is stopping shots. Is this type of information even legal to use during a game?

All of that is predicated on the fact that the coach isn't too stubborn or prideful to actually make use of the data.

I would think though that it's almost impossible to do all the "number crunching" during the game to the point you could make actual significant differences. Better would be to compile data from a single team under a coach and see if you notice "tendencies". The book is out on our powerplay, we the fans know it, other teams know it, the only one who doesn't seem to know it is RBA. Similar with our quantity > quality shooting policy. It's why we always slam face-first into a wall somewhere in the playoffs. In-game would be difficult to get sufficient data that your eyes can't already tell you (i.e. clogging up lanes, standing up blue-line, zone entries vs dump & chase), but pre-game would be huge. Have your "game plan" that you always do and if it hits a wall, use the data to help you try and change things up slightly to counter the other teams tendencies.

 

[Navin R Slavin]

I get that you were being tongue and cheek here but I'm actually surprised that we're not using more real time stats to make coaching decisions. I believe that statistics would make for a more interesting game, provided the players listen, there would be more scoring opportunities, more breakaways due to noticing tendencies, and higher scoring games due to better information on how a goalie is stopping shots. Is this type of information even legal to use during a game?

All of that is predicated on the fact that the coach isn't too stubborn or prideful to actually make use of the data.

In the PPS bit, I'm actually not bring facetious at all. The data is not the differentiator; the value is derived entirely from the appropriate filtering of the data. The more data that's collected, the truer that is. And when you're hiring data scientists to run your org, they know the end game: providing your humans with the simplest, most appropriate action at the exact right time.

I think that Brindy is actually a perfect coach for this. He doesn't need to be a perfect X and O guy. His greatest value is getting guys to buy in, so that they do exactly what the team requires of them: nothing more, and nothing less.

We have no idea what capabilities Tulsky can actually bring during a game, and he damned sure won't tell us. But we can make some really interesting guesses, based on who he recruits, and how. And as the state of the art evolves, which it absolutely will, I find it highly unlikely that anyone will be in front of him.

Fascinating times, these is.

 

[tarheelhockey]

there would be more scoring opportunities, more breakaways due to noticing tendencies, and higher scoring games due to better information on how a goalie is stopping shots.

Color me skeptical. Offense in this game is largely created by exploiting defensive mistakes. Giving coaches a bunch of fuel to eliminate mistakes sounds like a fast track to the next Dead Puck era.

 

[CanesFanBudMan]

Color me skeptical. Offense in this game is largely created by exploiting defensive mistakes. Giving coaches a bunch of fuel to eliminate mistakes sounds like a fast track to the next Dead Puck era.

The dead puck era is underrated

 

[Navin R Slavin]

Color me skeptical. Offense in this game is largely created by exploiting defensive mistakes. Giving coaches a bunch of fuel to eliminate mistakes sounds like a fast track to the next Dead Puck era.

Offense is also created by *causing* defensive mistakes, which is one of the reasons the Canes have invested so much in speed and play such an aggressive forecheck -- which is the opposite of the zero forechecker / left wing lock / neutral zone trap that was so dominant in the last dead puck era, right?

 

[tarheelhockey]

Offense is also created by *causing* defensive mistakes, which is one of the reasons the Canes have invested so much in speed and play such an aggressive forecheck -- which is the opposite of the zero forechecker / left wing lock / neutral zone trap that was so dominant in the last dead puck era, right?

Right*, but that’s just for puck acquisition. Once we have the puck, we’re incredibly systematized and not that much fun to watch. We are perfectly content to take 60 shots with a 5% chance of scoring, if that means we get 3 goals and can hope to get lucky for a 4th some nights. So our “product” is a shitload of low-danger point shots and half-screened wristers from the dot, which doesn’t scare anyone but moneyballs us to a win. Adding a bunch of AI influence is only going to make those margins thinner, as a computer model is always going to tell you to ignore 1:1 skill plays in favor of working the puck around for greater possession quality and shot volume.

Add AI on the defensive side, and I suspect what we get is a model to answer the question “So how can we make them score at 4% instead of 5%, so that our 6% shooting can all but ensure victory?”. So we’ll see even more disciplined defenses, less chasing the puck, less attempts to body check, zero tolerance for losing structure. And much like in the DPE, we gradually go from a 4-3 league to a 3-2 league.

Personally I’d be happy with a dramatic reduction in the amount of coaching. The game is most fun to watch when players are able to be creative and take risks. We tolerate the moneyball style because it’s winning us games, but I would not want to see an entire league like the Canes, let alone one even more driven by playing tiny margins.


* the change in eras from Neutral Zone Clot to Forecheck Chaos is mainly down to rule changes that opened up the neutral zone and turned it into an area where teams gather speed rather than losing speed.

 

[To Be Determined]

The stats are probably nothing novel. It's finding the data and putting it together that is incredibly complicated. Once you have the player speed and movement data accessible with traditional NHL stats, the modeling isn't a bad process.

Getting all that together in a coherent set? I'm impressed.

honestly, the most impressive thing in the tulsky article to me is that he possesses the type of intelligence that allows him to look at a data set and determine the type of "real world" problems that sort of align with the type of number crunching needed to glean useful information. then finding someone doing that work. the 'look for someone making an auto-drive sub' or whatever is genuinely insanely brilliant in a tesla (nikola, not the other guy) way.

 

[tarheelhockey]

honestly, the most impressive thing in the tulsky article to me is that he possesses the type of intelligence that allows him to look at a data set and determine the type of "real world" problems that sort of align with the type of number crunching needed to glean useful information. then finding someone doing that work. the 'look for someone making an auto-drive sub' or whatever is genuinely insanely brilliant in a tesla (nikola, not the other guy) way.

To me it’s a nod to his background. His career path through Berkeley, the Naval Research Lab, and Silicon Valley undoubtedly gave him a ton of exposure to the university research enterprise system, so he understands how tech companies sponsor faculty research; and if it’s successful, that they target those scientists for hire as young talent (or at least keep them on a sort of retainer by continuing to fund their work). He’s bringing that thought process over to the NHL.

Not to make this political, but when you see the headlines every day where NIH and NSF labs are being gutted, this is the death of the system that Tulsky came out of. 99% of it is focused on making the medicine that will save your life or the corn variety that will feed your children, not hockey.

 

[To Be Determined]

To me it’s a nod to his background. His career path through Berkeley, the Naval Research Lab, and Silicon Valley undoubtedly gave him a ton of exposure to the university research enterprise system, so he understands how tech companies sponsor faculty research; and if it’s successful, that they target those scientists for hire as young talent (or at least keep them on a sort of retainer by continuing to fund their work). He’s bringing that thought process over to the NHL.

Not to make this political, but when you see the headlines every day where NIH and NSF labs are being gutted, this is the death of the system that Tulsky came out of. 99% of it is focused on making the medicine that will save your life or the corn variety that will feed your children, not hockey.

absolutely. at the same time, most of the phd types i've known have been the 'inch wide, mile deep' folks in their fields. i guess stats is like the green cascading text in the matrix?

 

[Chrispy]

absolutely. at the same time, most of the phd types i've known have been the 'inch wide, mile deep' folks in their fields. i guess stats is like the green cascading text in the matrix?

Yes and no. I am capable of doing stats on any type of data, but if I don't understand what I'm analyzing I will inevitably end up in one of two places:

1) Reach a conclusion that is painfully obvious to anyone with a pulse (Eric, have the Hurricanes considered scoring more goals?)

2) Reach a conclusion that is clearly wrong but correlated according to the data.

And the more types of data you put together, the more you need to understand all of them. That's where Tulsky's obsession with sports stats and hockey in particular helps him take these data scientists with experience in a type of data and work with them to understand what is important in this data, and what is important in the hockey stats.

 

[To Be Determined]

not so random question for the group (that i am entirely too lazy to look up)...

i bought tickets to game 5. i went through ticketmaster, and removed all the "certified resale" or whatever they are, in the fear that if the canes somehow sweep (i ain't buying it fingers in ears lalalalala) the original purchaser might get the credit to carry forward to round 2 instead of me.

since it is 'certified' and goes through ticketmaster, was i just being paranoid and they would have given me a credit? or maybe they just refund it and i'm mixed up with back when we had season tickets and playoff tickets purchased for non-existent games just turned into credit against next years' season pack?

i assume for true secondary markets like stubhub or whatever the kids use these days, if you buy a game with the potential not to be played, you're just stuck? at the very best maybe a refund but no credit for the future?

edit - and yeah, i realize a refund is basically a credit that just goes back into your bank account to use again. i've talked myself into thinking i'm definitely mixing myself up with the old "carry unused ticket $$$ forward to next season' thing.

 

[Navin R Slavin]

...our “product” is a shitload of low-danger point shots and half-screened wristers from the dot...

Far be it from me to tell you what kind of hockey you like, but the Canes were 5th in the league this season in high danger shots for, with relatively modest offensive talent compared to the rest of the league.

And respectfully, I don't think you, or I, or anybody outside of that room, have any idea what specific insights Tulsky might currently be gleaning. These are the sorts of insights that only become evident in hindsight.

 

[tarheelhockey]

Far be it from me to tell you what kind of hockey you like, but the Canes were 5th in the league this season in high danger shots for, with relatively modest offensive talent compared to the rest of the league.

That’s kind of underselling how much talent the Canes have. Yes our top line talent is modest, but we have legit 20-goal guys on our third line and guys like Ghost and Orlov on our depth pairs. We’re a legit top-10 talented team in the league. The brilliance of the Tulsky-RBA collab is that we get that extra little nudge from about 10th to 5th, passing over teams that have high end talent but lesser depth.

But the way it happens is important. A lot of those high danger shots aren’t exactly an offensive machine at work, they’re the result of quick goals off the forecheck. Those are goals the other team isn’t scoring, and it’s a zero-sum game to grind the opponent down and break them mentally, to the point that they’re no longer even getting over the blue line. If you’re looking at the league as a whole unit, the Canes are doing more to reduce scoring than to raise it. Which is fine, as a Canes fan.

And respectfully, I don't think you, or I, or anybody outside of that room, have any idea what specific insights Tulsky might currently be gleaning. These are the sorts of insights that only become evident in hindsight.

Yeah, but history rhymes. In hockey, time and again we’ve seen a clear pattern where increased structure helps the defense. When expansion waters down the talent level, to the point that some players simply can’t keep up with others, scoring goes up. When the league gets sick of low scoring games and throws a rule-change curveball, scoring goes up. Then the coaches adjust to the new normal and scoring settles back down again. That’s been really consistent for a really long time.

This works differently in basketball, where the analytics say shoot a ton of 3s and win 150-140. Or in football where the analytics might say to forget about a running back and go five wide. In hockey, the numbers say you’ll win more games by playing extremely slim margins with shooting and save percentage, meaning a disciplined shot-suppressing game is the winning strategy. It’s the technically correct answer to the question. We’ve seen that in the team the Canes have built to execute Tulsky’s vision. The question is do we really want to see a league 32 teams just like the Canes, shot-surpressing each other to death? That sounds like it would kind of suck tbh.

 

[AD Skinner]

This doesn’t have anything to do with anything but apparently the boards are now letting you react with a leafs logo or senators logo but still no jokinen face, which is the exact type of situation I would want to use the jokinen face to react to

 

[Bub]

This doesn’t have anything to do with anything but apparently the boards are now letting you react with a leafs logo or senators logo but still no jokinen face, which is the exact type of situation I would want to use the jokinen face to react to

Panthers and Lightning now too.

Waiting for a California Golden Seals one or the KC Scouts. "Old time hockey!"

EDIT: Angy Jokinen is back!

 

[cptjeff]

"Hey, nerdy hockey kid. Want your name on the Stanley Cup?"

Candidate A: "I think you need more size in front on your power play."

Candidate B: "I think you need to find a better PPQB."

Candidate C: "Special teams (i.e. power play and penalty kill) situations play an outsized role in determining the outcome of ice hockey games. Yet, quantitative methods for characterizing special teams tactics are limited. This work focuses on team structure and player deployment during in-zone special teams possessions. Leveraging player and puck tracking data from the National Hockey League (NHL), a framework is developed for describing player positioning during 5-on-4 power play and 4-on-5 penalty kill possessions. More specifically, player roles are defined directly from the player tracking data using non-negative matrix factorization, and every player is allocated a unique role at every frame of tracking data by solving a linear assignment problem. Team formations naturally arise through the combination of roles occupied in a frame. Roles that vary on a per-frame basis allow for a fine-grained analysis of team structure. This property of the roles-based representation is used to group together similar power play possessions using latent Dirichlet allocation, a topic modelling technique. The concept of assignments, which remain constant over an entire possession, is also introduced. Assignments provide a more stable measure of player positioning, which may be preferable when assessing deployment over longer periods of time. Also, you have to shoot the damn puck once in a while."

Fixed.

 

[Derailed75]

Panthers and Lightning now too.

Waiting for a California Golden Seals one or the KC Scouts. "Old time hockey!"

EDIT: Angy Jokinen is back!

If we don't get a Cleveland Barons react I'm gonna be sad