The Advanced Stats Thread Episode VII: An Ode to the Sanity of Silverfish

[Ola]

Still, I do think its a bit risky to look too much at sv%.

 

[Mac n Gs]

Still, I do think its a bit risky to look too much at sv%.

I don't think anyone truly does anymore with metrics out there like Goals Saved Above Expectation (GSAE) and xFSv%. Goalie metrics have been improving a lot behind the scenes, but the mainstream media still uses GAA and raw sv% as their indicators of goalie success.

 

[Ola]

I don't think anyone truly does anymore with metrics out there like Goals Saved Above Expectation (GSAE) and xFSv%. Goalie metrics have been improving a lot behind the scenes, but the mainstream media still uses GAA and raw sv% as their indicators of goalie success.

Still, I feel that the result of those metrics very much resemble sv% if you get what I mean and is to a large extent adjusted sv%. Shot by location is also a very rough metric to adjust by.

You often need some kind of breakdown, the defense must face a little havoc, for a goal to be scored. Consistency is very important, just to know what to expect. It can be easier to play for a team that always stink than a team that stinks every other night. Do the shots come from the same places night after night? When a D is beaten, is it the same D that is slow and cause 10 breakaways per season or does everyone take turn? If its the same guy you start to prepare before it happens, if not its more of a surprise. There are soooo many team related things that play a part in sv% or sv% adjusted stats.

Look, all I know is this, looking at the 31 teams in this league -- can you adjust so that you get to compare apples to apples looking at sv% or sv% adjusted stats? Its going to be very hard. Its a team game.

 

[East Coast Bias]

Looking for some help. For those that know Python - what's the best site/book/way to self-teach it?

I have some time coming up where I'll be traveling for work and sitting in hotels, so I want to learn something new.

My background is more on the analysis side, less programming. Fwiw.

Any info/suggestions is appreciated.

 

[silverfish]

Looking for some help. For those that know Python - what's the best site/book/way to self-teach it?

I have some time coming up where I'll be traveling for work and sitting in hotels, so I want to learn something new.

My background is more on the analysis side, less programming. Fwiw.

Any info/suggestions is appreciated.

So you'll be traveling, and instead of exploring, you're going to sit in your hotel room, and learn Python?

WELCOME, BROTHER!

I think people like this: Python Data Science Handbook | Python Data Science Handbook. But my foray into Python ended as quickly as it started mostly because there's no Python IDE that touches RStudio.

 

[East Coast Bias]

So you'll be traveling, and instead of exploring, you're going to sit in your hotel room, and learn Python?

WELCOME, BROTHER!

I think people like this: Python Data Science Handbook | Python Data Science Handbook. But my foray into Python ended as quickly as it started mostly because there's no Python IDE that touches RStudio.

HA it's to a boring city in the heartland which I'll leave anonymous as to not offend others. The place sucks. Only so many times you can walk across the plaza to chili's for 2 for 1s.

Thanks for this. So you started with Python, but shifted to R instead? I hear Python is easier to learn?

I understand SQL on a high level, and ETLs in general - I just never got into programming any of it. Knew enough to talk to programmers and customers alike. Want to learn it more though.

 

[silverfish]

HA it's to a boring city in the heartland which I'll leave anonymous as to not offend others. The place sucks. Only so many times you can walk across the plaza to chili's for 2 for 1s.

Thanks for this. So you started with Python, but shifted to R instead? I hear Python is easier to learn?

I understand SQL on a high level, and ETLs in general - I just never got into programming any of it. Knew enough to talk to programmers and customers alike. Want to learn it more though.

I was on R. Attempted to shift to Python because rumor is Excel will add Python compatibility and my industry is still using mostly Excel. I may pick up again in future, but no need right now.

Chili's is dope. I mean not really, but ya know.

 

[silverfish]

Somebody really needs to do a deep-dive analysis on regular season Tonathan vs post-season Tonathan because at this point this shit is stupid.

 

[Irishguy42]

Somebody really needs to do a deep-dive analysis on regular season Tonathan vs post-season Tonathan because at this point this **** is stupid.

The only deep-dive you really need is:

Regular season JTM - Apparent god-send
Post season JTM - LOL

 

[Synergy27]

Looking for some help. For those that know Python - what's the best site/book/way to self-teach it?

I have some time coming up where I'll be traveling for work and sitting in hotels, so I want to learn something new.

My background is more on the analysis side, less programming. Fwiw.

Any info/suggestions is appreciated.

I think this is the right answer:

Learn Python the Hard Way

For R I recommend the Johns Hopkins course on Coursera but I took it about 4 years ago and I bet it's not free anymore.

 

[Ola]

I take the analytics team, just like 10/12 of the playoffs series.

Hehe, maybe it would work. With the technology used by the IIHF at the WCHs you will start to get big data that seriously could start to push it if hooked up to a Watson.

NHL players are so complete and smart nowadays. It might not be that its as important to "handpick" a lineup. Looking at the other side of the spectrum, if you like coach 12 y/o's, one roster can literary lose by 10 goals against another roster one night and then win by 10 goals against the same roster two weeks down the road. I saw a team of 14 y/o lose 16-4 last fall and then win against the same team 6-0 just a month later. If your play don't work it don't work, and the marginals are really small, but its not as obvious in the NHL. Even if I don't think the world of all NHL coaches, they of course get the basics right. With kids the contrast is so obvious. Forechecking is instinctive and somewhat easy, getting up ice is hard. If a team isn't prepared to handle a situation, execute a break-out play that works against a fairly good defense, something like that, it will cost you every single shift. Even if you don't care that much or at all about winning when coaching kids, normally you really have to be on your toes because if you don't get it right the coming 3x20 can be such a long time... Hockey is a very unforgiving game in that sense.

The always and everywhere -- to some degree -- existent disconnect between fans and coaches is definitely a product of that. The coach is soooo focused on getting everything to work. Even in the NHL, you don't have to try to sabotage things that much to lose by 10-0 or more, every game. Players differs a lot. Will all lines have players with the presence of mind to think of X, Y and Z, every shift all game? And its not just 3 things, its easily 30 things that isn't a given. A schoolbook example: if a defender goes deep into the attacking zone a forward is supposed to drop down and take his place. That will only happen if someone of the forwards is constantly aware if a defender goes deep or not. If it doesn't work you will get a 3 on 1 or 2 on 1 against you. So when you put together a line. and the entire lineup ultimately, you have to consider this. Do I have a guy on each line that always will have the presence of mind to drop down and take the defenders spot?

And its the same in so many other areas. If you have 3 different set of break-outs, it means that you also have 18 skaters that must do the exact right thing depending on which breakout that is chosen on the ice, what the other players on the ice do and how the defensive team lines up. Its the same on the forecheck, you have to adjust and make the right decision all the time all over. And besides the decision making -- you also have the ability part that is just as important.

When creating a line-up -- you have to ask yourself a whole bible of questions and feel comfortable with the sum of all positive answers, everything can't be perfect.

To further explain this, lets look at a more realistic NHL example: lets say that you have a D that sucks at moving the puck, lets call him Dan Girardison. You play NJD and they want to be smart and go out of their way to disturb things for you, so before the game they say listen you know our gameplan, follow it, BUT every single time Dan Girardison goes into the corner to fetch a puck, you shall not pressure him but instead take away all his passing options. And try to dump the puck as often as possible in his corner. We want him to be forced to skate the puck up ice as often as possible, he will wind up tripping over his own stick 1/3 of the time. On NJDs side, you need players on the ice that can buy into the concept and is good at it, if that is your game-plan. Its a disaster if your forwards back down but all of a sudden a D freaks out and tries to step up on DG at the redline leaving a forward open. On NYRs side, you must have guys on the ice that knows how to act in this situation. A solution can be to go really deep with two other forwards and curve towards Girardi, the defending team cannot really drop with two defending players down towards the net in the attacking zone if a third player has the puck and open ice in front of him. Most likely you will end up in a situation were a forward, probably the center, have to take a very big responsibility reliefing Girardi of the puck and then being able to skate it up ice.

The question a coach have to ask himself in relation to the above is -- can I play Girardi despite him potentially being exposed like this? If I do, can I surround him with players that can solve it if we face this strategy?

I can guarantee that when a coach plays a "4th lineer" on a "1st line" -- and a bunch of fans goes is he nuts, can't he see that the plug in question sucks at hockey? -- the reason for it is not that the coach can't see what the fan sees, its because the coach have asked himself questions like the above and come to the conclusion -- right or wrong -- that certain requirements were needed and that the plug provided those abilities.

The NHL game is very complex: Teams spends hours and hours coming up with and adjusting to strategies, and further hours and hours trying to implement everything. Its like directing a ballet. Its not good if someone don't know what to do or someone can't make a lift you want that someone to do.

My point is just, there is a tremendous amount of 'hands on work' when building a hockey team. Its not about finding the 18 best skaters and then just put them into a line-up starting filling in the best names at the top of the line-up. There is no such thing as the "best" player. I am exaggerating a ton of course, but somewhat think of it as building an engine for a car. You cannot go l like the cylinders best of all engine parts so the only part I am ordering is 100 cylinders, it will make a heck of an engine... Hockey is about scoring goals, I am getting 18 scorers.

But on the other side of the spectrum, hockey players are becoming more complete. They are so well educated. There are much fewer role players. I can guarantee that you 20 years ago not even remotely just could have put a random roster on the ice, of players distinguishing themselves in certain isolated areas. You could have a Cicarelli on a line, but if you had 2 it would have been good night.

I just think that I think we would benefit a bit from understanding the context and complexity of things better, a lot of our arguments comes out of things being oversimplified!

 

[Mac n Gs]

@Ola, you have no idea how badly I want that IIHF system to be installed in the NHL, if it isn't already, and to have some of that data be publicly available. I'd also love to know what AV's sophisticated stats package entailed.

Anywho, I think it was Micah or Garret Hohl, but one of them actually had a post about how it's not so crazy that the Caps might beat the Bolts. I'll try to find it and share it in here.

 

[silverfish]

The only deep-dive you really need is:

Regular season JTM - Apparent god-send
Post season JTM - LOL

It really makes no ****ing sense, though. I don't really have the time to dive in right now, but I want to. All my free time that I actually dedicate to analysis is going to be for the similarity model and NWHL work.

@Mac n Gs step up

 

[1Knee1T]

It was Micah, essentially explaining why his model chose Washington over Tampa. I shared his tweet thread a few pages back. I'll dig it back up.

Edit: Here it is.

 

[Mac n Gs]

My guess: he doesn’t get shots off in tight in the playoffs. He’s good at getting near the net and circles in the regular season, but he’s not good around the net when teams collapse more and can get away with more in the playoffs.

He was a poor shot generator, especially off the rush, so he essentially gets neutered every playoffs. His xFsh% spikes every playoffs.

 

[silverfish]

My guess: he doesn’t get shots off in tight in the playoffs. He’s good at getting near the net and circles in the regular season, but he’s not good around the net when teams collapse more and can get away with more in the playoffs.

He was a poor shot generator, especially off the rush, so he essentially gets neutered every playoffs. His xFsh% spikes every playoffs.

I want to see:

Shotmap comparisons
iCF60 comparisons
ixFSh% comparisons
ixG60 comparisons

% time on PP comparisons.

Gimme. gimme. gimme.

 

[Mac n Gs]

I want to see:

Shotmap comparisons
iCF60 comparisons
ixFSh% comparisons
ixG60 comparisons

% time on PP comparisons.

Gimme. gimme. gimme.

If you know where I can get the heatmaps, I’ll do it. I’ll do the rest later today once I’m done taking care of some work

 

[silverfish]

If you know where I can get the heatmaps, I’ll do it. I’ll do the rest later today once I’m done taking care of some work

Would have to scrape and compile yourself, I don't think any site has the compatibility to compare reg season to playoffs by player. I'm not sure it has any true usefulness outside of what I want to do with JT.

I can do it, I just don't have the time

 

[Ola]

It was Micah, essentially explaining why his model chose Washington over Tampa. I shared his tweet thread a few pages back. I'll dig it back up.

Edit: Here it is.

How many times must the series be replayed to verify the result, 100?

Predicting the outcome of a PO series, that is so close that it goes to 7, must be like predicting if a player will score on a SO or not.

The last decade I've felt that the Caps was dangerous, sometimes more so than others. They have lost many G7s and what not, the marginals are very small. Some of those have been coin tosses, them finally going to the finals is IMO not that surprising.

 

[1Knee1T]

How many times must the series be replayed to verify the result, 100?

Predicting the outcome of a PO series, that is so close that it goes to 7, must be like predicting if a player will score on a SO or not.

The last decade I've felt that the Caps was dangerous, sometimes more so than others. They have lost many G7s and what not, the marginals are very small. Some of those have been coin tosses, them finally going to the finals is IMO not that surprising.

According to Micah you would need to play it out a "gargantuan number of times"

 

[GeorgeKaplan]

Recent Rangers south had a rough ECF

 

[YoSoyLalo]

Gorton said Quinn is very open to analytics use and said webuse analytics too

So basically what this boils down to is Vally needs to come up with better stats

 

[silverfish]

Gorton said Quinn is very open to analytics use and said webuse analytics too

So basically what this boils down to is Vally needs to come up with better stats

NO SHOT

 

[Ola]

Recent Rangers south had a rough ECF

How is this counted?

 

[GeorgeKaplan]

How is this counted?

Dimitri Filipovic counted and kept track of all the zone entries/exits himself, those numbers are just for the conference finals, but he posted all the raw totals on twitter