Jump to content
The Official Site of the Montréal Canadiens
Canadiens de Montreal

The Great Stat Discussion


Recommended Posts

Here's a test graph which shows both 5-on-5 and all situations lines on the same plot:

20001-1452293990.png

Any thoughts? I'm worried it might be too busy. But I really like being able to see both at a glance.

If you're satisfied with functionality and want to make it more aesthetically appealing ... Keep in mind that this is being extra picky and only because you brought it up. Functionality is most important, I'm only making suggestions that may make it more presentable. I like what you've done. :)

1. Shrink the legend into a small overlay box and place it on the top left corner. (You posted one from another site that did this previously). That would cut down some of the "clutter".

2. A color change at "0" on the Y-axis would be cleaner and easier to read than a gradient.

3. Graph lines probably shouldn't disappear into the edge of the graph. You may want to add a bit of a margin there.

4. For running both lines on the same graph, the focus line should be evident, but not straight black. It takes all of the focus in contrast to the other colors. A good neutral color (such as a slate gray or a grayish blue) may work better. A dotted line for the secondary line would likely work well, if not it just needs to be a few shades lighter than the primary line. If either line has to be white, make the primary a couple of shades lower than stark white and the secondary line gray.

5. The goal markers could be smaller. Ideally, they shouldn't obscure the next event on the X-axis.

6. Avoid "stroke" effects, or borders on lettering (and lines) where a different color would serve the same purpose. It just looks more organic that way (rather than dropped on top like a fan-subbed video).

Again, only an attempt to help! :P

Link to comment
Share on other sites

How dare you try to help! :lol:

Making the legend better is definitely on the to-do list. It's way too big right now, and somewhat ugly.

2. A color change at "0" on the Y-axis would be cleaner and easier to read than a gradient.


This is what I did initially, and it felt too flat and generic to me:

20001-1452301421.png

Going forward, I'd ideally like something different, but I'm not sure what it'll be.

3. Graph lines probably shouldn't disappear into the edge of the graph. You may want to add a bit of a margin there.


This is going to be fixed immediately after I fix the scaling/size issue.

4. For running both lines on the same graph, the focus line should be evident, but not straight black. It takes all of the focus in contrast to the other colors. A good neutral color (such as a slate gray or a grayish blue) may work better. A dotted line for the secondary line would likely work well, if not it just needs to be a few shades lighter than the primary line. If either line has to be white, make the primary a couple of shades lower than stark white and the secondary line gray.

This is the result of my current limited knowledge of R. The line plotting function's dotted line looks rather poor on these right now. I'd like something different, but right now I think I'm going to have to settle with the current lines colour arrangement.

5. The goal markers could be smaller. Ideally, they shouldn't obscure the next event on the X-axis.


It'll be hard to get the markers small enough to avoid obscuring every following event. Here's a reduced size version, though:

20001-1452301842.png

I definitely like that better, myself.

As for the text outlining, that's because I had issues with the text being really unreadable without it. I'm going to have to play around with it some more.

Link to comment
Share on other sites

How dare you try to help! :lol:

Making the legend better is definitely on the to-do list. It's way too big right now, and somewhat ugly.

This is what I did initially, and it felt too flat and generic to me:

20001-1452301421.png

Going forward, I'd ideally like something different, but I'm not sure what it'll be.

This is going to be fixed immediately after I fix the scaling/size issue.

This is the result of my current limited knowledge of R. The line plotting function's dotted line looks rather poor on these right now. I'd like something different, but right now I think I'm going to have to settle with the current lines colour arrangement.

It'll be hard to get the markers small enough to avoid obscuring every following event. Here's a reduced size version, though:

20001-1452301842.png

I definitely like that better, myself.

As for the text outlining, that's because I had issues with the text being really unreadable without it. I'm going to have to play around with it some more.

Smaller goal markers is an improvement. I just didn't know how concerned you are with little details and whether more suggestions are just a pain in the bottom or appreciated. Some people don't like criticism, even if it is constructive. ;)

As for the flat background, I see what you mean. Perhaps a lighter color or subtle texture instead?

Is it just a static background image or are you painting it dynamically?

Link to comment
Share on other sites

Smaller goal markers is an improvement. I just didn't know how concerned you are with little details and whether more suggestions are just a pain in the bottom or appreciated. Some people don't like criticism, even if it is constructive. ;)

As for the flat background, I see what you mean. Perhaps a lighter color or subtle texture instead?

Is it just a static background image or are you painting it dynamically?

Oh, of course. Thanks a ton, BTW. I want to make it as good as possible. I doubt I'm currently capable of this, but an ultimate goal would be to make this the best of breed in possession charts. So I'm looking to fix everything.

The backgrounds are procedural polygons, based on some code I found on Stack Overflow:

nslices = 255
colsT <- colorRampPalette(colors = c("white",MTLC))(nslices)
ys <- seq(0,Yscale,len = nslices + 1)
for (i in 1:nslices) {
polygon(c(0,0,Xscale,Yscale),c(ys[i],Xscale+32000,ys[i],ys[i]),col=colsT[i],border=NA)
}
Link to comment
Share on other sites

Oh, of course. Thanks a ton, BTW. I want to make it as good as possible. I doubt I'm currently capable of this, but an ultimate goal would be to make this the best of breed in possession charts. So I'm looking to fix everything.

The backgrounds are procedural polygons, based on some code I found on Stack Overflow:

nslices = 255
colsT <- colorRampPalette(colors = c("white",MTLC))(nslices)
ys <- seq(0,Yscale,len = nslices + 1)
for (i in 1:nslices) {
polygon(c(0,0,Xscale,Yscale),c(ys[i],Xscale+32000,ys[i],ys[i]),col=colsT[i],border=NA)
}

I made a quick mock-up before seeing this, but here's a general idea anyway, maybe you can take something away from it.

If you want something top of the line and professional looking, you'll have to go a step beyond. I'd be thinking along the lines of an ice surface type texture, including logos, rounded edges, etc.

5UJHW6S.png

Link to comment
Share on other sites

I made a quick mock-up before seeing this, but here's a general idea anyway, maybe you can take something away from it.

If you want something top of the line and professional looking, you'll have to go a step beyond. I'd be thinking along the lines of an ice surface type texture, including logos, rounded edges, etc.

5UJHW6S.png

I definitely like the feel of that more than the current one. I'll be implementing that legend ASAP. Logos present a potential copyright issue, which I'd like to avoid. I like pretty much everything else, too. I'll look into implementing a PNG background and multiplying the plotted components. A next generation version might even be functional for the game tomorrow. Thanks! :D

Link to comment
Share on other sites

I definitely like the feel of that more than the current one. I'll be implementing that legend ASAP. Logos present a potential copyright issue, which I'd like to avoid. I like pretty much everything else, too. I'll look into implementing a PNG background and multiplying the plotted components. A next generation version might even be functional for the game tomorrow. Thanks! :D

No problem. Happy to help.

Hoping I can make the game tomorrow, so I'll see any changes then.

I think the biggest takeaway between the two, is that many of your original elements are rather stark. What should stand out is the graph lines and goals, as that is the main information you wish to display. The rest can be tuned down or made more subtle/simplistic.

As for logos, you could also try a pattern of team names in the background as well. It would serve as visual appeal and to identify each half, so you wouldn't necessarily have to add the team names either. Just another thing you can consider.

Link to comment
Share on other sites

I present version 2.0. Actually, two of them:

20001-1452324141.png

20001-1452324096.png

I'm not sure which of the two I like more. The margin legend variant is a bit harder to read, but the inset legend version could (and let's hope, regularly will) obscure the data.

I'd say the 2nd one for that reason.

Looks much better now imo.

I'd suggest removing the white from the black graph line (stroke) and the goal markers. Other than that, looks good to me.

(Unless you want to add some texture to the background, but I don't think it's necessary.)

Link to comment
Share on other sites

I'd suggest removing the white from the black graph line (stroke) and the goal markers. Other than that, looks good to me.

(Unless you want to add some texture to the background, but I don't think it's necessary.)

Yeah, that looks better:

20001-1452364930.png

I'll look into texture later, but this is way solid now. Compared to how it looked not 24 hours ago, it's night and day. Thanks a million!

Link to comment
Share on other sites

Yeah, that looks better:

20001-1452364930.png

I'll look into texture later, but this is way solid now. Compared to how it looked not 24 hours ago, it's night and day. Thanks a million!

No problem, glad you're happy with it.

Your secondary line is still hitting the border, not sure if that was intentional.

There appears to be a random vertical line attached to the first Leafs goal (which you probably noticed anyway).

I might prefer a slight drop shadow on the white line instead of an emboss, but that's minor.

I think everything else looks better as well. Much cleaner and easy to read. Looks pretty, nice job! :D

Link to comment
Share on other sites

No problem, glad you're happy with it.

Your secondary line is still hitting the border, not sure if that was intentional.

There appears to be a random vertical line attached to the first Leafs goal (which you probably noticed anyway).

I might prefer a slight drop shadow on the white line instead of an emboss, but that's minor.

I think everything else looks better as well. Much cleaner and easy to read. Looks pretty, nice job! :D

Yeah, the lines are still being plotted over the outline because I'm lazy. :lol:

And actually, that line around Toronto's first goal is a power play polygon that the previous setup didn't detect. Gallagher took a penalty 13 seconds into the second period, and van Riemsdyk scored 6 game seconds after.

Link to comment
Share on other sites

@IneffectiveMath of HockeyViz has very generously allowed me to use his score adjustment formula in the Corsi graph!

2015020616-1452447927.png

Can you share the formula and/or show one with and one without the score adjustment?

(Would like to know and see what the difference is and the reasons for the difference, otherwise it's just pretty lines :P)

The Pittsburgh goals also disappear. Like I said, you may need to fine tune some colors. ;)

Link to comment
Share on other sites

Can you share the formula and/or show one with and one without the score adjustment?

(Would like to know and see what the difference is and the reasons for the difference, otherwise it's just pretty lines :P)

The Pittsburgh goals also disappear. Like I said, you may need to fine tune some colors. ;)

Yeah, he's got an article about it:

http://hockeyviz.com/txt/shiftsArticle/senstats.html

Those coefficients are older, here's the set I'm using:

http://hockeyviz.com/txt/saf-coeffs.json

And here's another article:

http://hockey-graphs.com/2014/11/13/adjusted-possession-measures/

Link to comment
Share on other sites

Is there data available on quality and type of shot, and has any pertinent information been discovered there?

I'm not talking about scoring chances or shot location, but the quality of a shot in and of itself.

- Screened shots

- Slap shots

- Wrist shots

- One timers

- High and low shots

- Shot speed

- etc.

I'm wondering because the value of corsi only related to quantity and not quality, while shot quality is currently only measured by location.

Taking a weak shot at the goalies chest shouldn't be considered a quality scoring chance and it seems to me a large proportion of our shots, despite how many we are getting, have little chance to beat a goaltender.

Link to comment
Share on other sites

I'm not aware of any published shot quality system that has been shown to be comprehensively useful in the sense you're talking about. The RTSS data include (as you mentioned) shot location, and also type (wrist, one timer, etc). These are inputted on the spot by the individuals doing the recording, I believe with a touch screen. WOI has a very comprehensive rink bias and error correlation system which they've published about (I'll look for links in a bit), and others have worked with the RTSS data as well. Ideally, though, we'll want player and puck tracking, with photogrammetry, RFID, something else, or a combination of all of it. Proprietary photogrammetric player tracking systems do currently exist, but they're mostly used for broadcast purposes. MLB is on the forefront of this, and I wouldn't be surprised if most of what ends up in hockey will come from them.

Using location to delineate scoring chances is far from ideal, but it's probably the least sketchy method we have right now as amateurs. WOI also tries to categorize shots as either on the rush or from rebounds by looking at inter-event timing, which is something I'd like to get into.

Link to comment
Share on other sites

Thanks! I wasn't sure but I did doubt it currently was available. You'd have to watch the games and apply subjective judgement without some advanced tracking. I know someone was doing that for something with zone entries with and without possession.

Shot type isn't ideal either, but if there are numbers available which suggest certain types of shots have sufficiently different probability of resulting in goals, it might be possible to apply those adjustments to Corsi. If shot location or some of the other factors could be applied as well, Corsi might be able to be judged alongside a modified stat.

We know outshooting an opponent increases the likelihood of outscoring them, but that can be significantly reduced for a team taking shooting opportunities with a low probability of success. It would be nice if there were a way to measure that, but it may be something that has to wait for technology to catch up after all.

Link to comment
Share on other sites

Yeah, the current generation of data aren't good enough to derive a ton of meaning from, shot-quality wise. It seems tempting to assume the RTSS location data error is small enough to allow you to ignore long shots from the point, or such. But there's no way to determine if there was traffic in front of the net, and so on. While it's true you're binning Ovechkin snipes together with centre ice slapshots at the goalie, it has to be remembered:

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

Guest
This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...