Player Specific Tracking Data

[primetime]

This. If the teams are smart, they'll apply machine learning by feeding an algorithm the data and (subjective) results. For example, label each play as positive or negative and construct an algorithm to maximize positive plays with that data.

Machine learning is a nice buzzword but there's little evidence it produces anything more predictive than manual hypothesis testing and model building. I have a neural network. It doesn't do ****.

 

[Kontradiction]

GFY.
Grow up, ***hole

AWESOME sense of humor. You realize I was breaking your balls, correct?

 

[XLIX]

AWESOME sense of humor. You realize I was breaking your balls, correct?

So was I. So what’s the problem?

 

[jah]

Machine learning is a nice buzzword but there's little evidence it produces anything more predictive than manual hypothesis testing and model building. I have a neural network. It doesn't do ****.

Computer vision has been absolutely ****ing revolutionized, and the implications of that are incredibly far reaching as well as applicable to pretty much any other type of data. That industry is what I would point to as a demonstration of the power of neural networks.

 

[primetime]

Computer vision has been absolutely ****ing revolutionized, and the implications of that are incredibly far reaching as well as applicable to pretty much any other type of data. That industry is what I would point to as a demonstration of the power of neural networks.

What industry? Computer vision is pretty negligible in terms of productivity, and has mostly been driven by force feeding cheap (mechanical turk) or free (all those verification tests you make) human labor into those neural nets.

For what it's worth, I think reports of revolutions in IT are ridiculously overplayed at this point. The market agrees with me, considering private investment in computing is at its lowest rate since the early 1980s and in terms of absolute investment remains well below the highs of the late 1990s and mid 2000s.

In any case, none of this has any bearing on its worth for football analysis, which seems remarkably limited. You'd almost need to recruit a ton of people to determine whether player X or player Y is more likely to make a play given their location on a given play. But even then I don't think lay people reading numbers would help much. Might have to attach it to actual footage then ask to feed it into the model.

But... like, why bother? You could have someone just watch the damn film and come to the same conclusion in a fraction of the time. You'd lose only the mystical voodoo of numbers.

 

[holyredeemer]

What industry? Computer vision is pretty negligible in terms of productivity, and has mostly been driven by force feeding cheap (mechanical turk) or free (all those verification tests you make) human labor into those neural nets.

For what it's worth, I think reports of revolutions in IT are ridiculously overplayed at this point. The market agrees with me, considering private investment in computing is at its lowest rate since the early 1980s and in terms of absolute investment remains well below the highs of the late 1990s and mid 2000s.

In any case, none of this has any bearing on its worth for football analysis, which seems remarkably limited.

Geeeeezzzz fellas. Take it somewhere else. This is a Football forum, not a goofball forum. Nerds!!!

(Edit-Sorry. Me lashing out during intelligent conversation is merely a coping mechanism to make my self feel more significant. Please, carry on.)

 

[primetime]

Geeeeezzzz fellas. Take it somewhere else. This is a Football forum, not a goofball forum. Nerds!!!

(Edit-Sorry. Me lashing out during intelligent conversation is merely a coping mechanism to make my self feel more significant. Please, carry on.)

You're the smarter one for not having an opinion on this ****.

 

[DarrylS]

It's gonna be a long offseason..

 

[luuked]

You could have someone just watch the damn film and come to the same conclusion in a fraction of the time. You'd lose only the mystical voodoo of numbers.

Because youe measurements will be (somewhat) normalized and not completely skewed by assumptions you need to make when watching tape.

At worst you can at least use the data to verify that the work your scouting dept is doing is not totally off.

 

[PatsFanInVa]

Saw the story too, was going to post it but the board was down

I like that "some teams" were ahead of others blah blah blah... and the opening of the floodgates was directed from on high (the infamous Competition Committee.)

"Ex Post Facto DataGate" in three... two... one...

But sirrously... in theory this should "level the playing field" in terms of preparation, just as making it illegal to cover receivers leveled the playing field in terms of wide receiver importance.

I don't know what sensors you get. The article I read said "sensors in their shoulder pads." You could have a tendency to lift a shoulder a certain way when you're about to do a certain move in line play, or when you're walking up to the line and your task is to blitz, not to drop back (or vice versa).

Hopefully we already have good software to process game tape with some representation of the data for the visualization pleasure of the dark lords of Foxborough

 

[PatsFanInVa]

Machine learning is a nice buzzword but there's little evidence it produces anything more predictive than manual hypothesis testing and model building. I have a neural network. It doesn't do ****.

Feed in the rule that this is play X, as run in the presence of defense Y.

Use all players at a position to decide what the characteristics are of all players in that play. Hell figure in clock/down/distance. Should analogous to saying "here are examples of how people write the letter A, now recognize the letter A"

Then say "hey neural network, you be the judge: what is this player behavior indicative of, for this particular player, in this situation?"

You should have a very efficient tool for diagnosing tells, augmenting a coach's or player's ability to say "I've noticed that Jones does X whenever he's going to do Y."

You should also have a very efficient tool for saying "This looks like a power sweep, but watch the tight end; If he does X, then Y."

Question is whether they'll make the streaming data available unprocessed in-game. Also whether they can mate the trackers to electrical shocks when the coach doesn't like what he sees in practice.*

*edit: Or optimally, if the tablet doesn't like what it sees in practice.

Google won't let me copy a picture of a tablet from Westworld to paste here. I do apologize.

 

[primetime]

Feed in the rule that this is play X, as run in the presence of defense Y.

Use all players at a position to decide what the characteristics are of all players in that play. Hell figure in clock/down/distance. Should analogous to saying "here are examples of how people write the letter A, now recognize the letter A"

Then say "hey neural network, you be the judge: what is this player behavior indicative of, for this particular player, in this situation?"

You should have a very efficient tool for diagnosing tells, augmenting a coach's or player's ability to say "I've noticed that Jones does X whenever he's going to do Y."

You should also have a very efficient tool for saying "This looks like a power sweep, but watch the tight end; If he does X, then Y."

Question is whether they'll make the streaming data available unprocessed in-game. Also whether they can mate the trackers to electrical shocks when the coach doesn't like what he sees in practice.*

*edit: Or optimally, if the tablet doesn't like what it sees in practice.

Google won't let me copy a picture of a tablet from Westworld to paste here. I do apologize.

I don't think this is any different than what most teams already do. But you're still leaning on eleven players to internalize it and make split second reads in the eight or so seconds between a team lining up and them snapping the ball.

 

[PatsFanInVa]

I don't think this is any different than what most teams already do. But you're still leaning on eleven players to internalize it and make split second reads in the eight or so seconds between a team lining up and them snapping the ball.

Well that's why the players themselves should be controlled by joysticks. But now you're just being utopian.

 

[primetime]

Well that's why the players themselves should be controlled by joysticks. But now you're just being utopian.

If only we could watch football games controlled by forums poster Chris Stevenson

 

[PatsFanInVa]

So let's consider it an open question for the time being. At the very least, I could see a "tendencies" algorithm developing to the point where it notices some things that the humans don't, but garbage in, garbage out. "A" sensor in each player's shoulder pads would have to be correlated to whatever else shows up on tape... why the torso moved. Other tells would not be evident, e.g., shoulders remain same level and orientation, but feet pointed slightly differently...? Certainly facial expressions would be very difficult to capture. Good thing too, what with the Tom-Looking-Off-The-Coverage trick.

I just figure you'd have at least one valuable tool, worst case.

Best case - probably limited by the limited data - would be that you let the machine write itself a nice little code that notices what you don't. Alpha Go and all that.