Blackbox modeling (Video)

Justin7 · 03-18-09, 08:40 AM

Part II

compaqDikk · 03-18-09, 08:47 AM

i came in here thinking you were gonna be modeling for us in black boxers. what a letdown

SlickFazzer · 03-18-09, 09:00 AM

Good info, thanks.

spongerat · 03-18-09, 10:51 AM

great as always justin. makes peoples "systems" look simplistic and foolish in comparison.

jjgold · 03-18-09, 11:16 AM

Well the simulation models are just as good as any other handicapping theory Solid and informative vids Did you notice Justin painted his window black? That is fukkin crazy. What is this guy into?

onthewhat · 03-18-09, 11:32 AM

Guy painted his fukkin prick

SlickFazzer · 03-18-09, 11:52 AM

Originally posted by onthewhat

Guy painted his fukkin prick

There is only one guy on the board who has ever painted his fukkin prick, he had paint all over his ass and prick.....

dwaechte · 03-18-09, 07:54 PM

Great stuff Justin. I think this is what a lot of the newer bettors around here had in mind for what they wanted from a video. We've seen a countless number of rec bettors come on the board and want to know how models are developed, and this is a nice quick, easy, summary of how to go about it.

Justin7 · 03-18-09, 11:36 PM

Originally posted by dwaechte

Great stuff Justin. I think this is what a lot of the newer bettors around here had in mind for what they wanted from a video. We've seen a countless number of rec bettors come on the board and want to know how models are developed, and this is a nice quick, easy, summary of how to go about it.

Thanks. It always surprises me that the more good information I give out, the less interested people are. The more basic (with no real meat), the more popular the videos are.

flyingillini · 03-18-09, 11:39 PM

I like The Greek lines up in the background.

Justin7 · 03-18-09, 11:49 PM

Originally posted by flyingillini

I like The Greek lines up in the background.

I was betting futures between scenes. I saw that, and panicked for a second. Could people see my account? Or my balance? I didn't want a repeat of earlier.

Fiasco · 03-18-09, 11:56 PM

Originally posted by Justin7

Thanks. It always surprises me that the more good information I give out, the less interested people are. The more basic (with no real meat), the more popular the videos are.

haha because gamblers are stupid and they don't want to learn anything...

they want to be spoon fed easy to comprehend material.

Good job though and keep it up

Wrecktangle · 03-19-09, 12:02 AM

Justin7: "It always surprises me that the more good information I give out, the less interested people are. The more basic (with no real meat), the more popular the videos are."

I once was surprised too. Nobody wanted to talk math, they just wanted my picks. But let's face it, people want action and hate math and all the work that goes into analysis. For a number of years I built sports models and didn't even bet on them...I just wanted to see if I could paper trade a sport to see if it was winnable. Modeling was always much more fun. I hate chasing down "the best line"...I'd much rather tweak up the sim to get me that extra 1/2 point than grub around differing books to find it.

Wrecktangle · 03-19-09, 12:16 AM

Oh, and by the way: those are good videos. Your point on out of sample testing branded you as a serious modeler, a point that few folks make. I disagree with a few points, but on the whole I'd rate them better than almost any I've seen on the topic...matter of fact, I can't recall having seen any (not on sports modeling, at least), so by definition they are the best I've ever seen.

pico · 03-19-09, 12:25 AM

economagic with sports data. solid videos. gambling aint easy

so im zach · 03-19-09, 12:55 AM

reno cool · 03-19-09, 03:42 AM

Since books have much more resources than bettors why is it that a statistical approach can beat them?
1. Because they intentionally skew the line to take advantage of public perception.
2. There's an inherent advantage one has betting into a line, over the one posting the line
3. something else.

pico · 03-19-09, 04:19 AM

Originally posted by reno cool

Since books have much more resources than bettors why is it that a statistical approach can beat them?
1. Because they intentionally skew the line to take advantage of public perception.
2. There's an inherent advantage one has betting into a line, over the one posting the line
3. something else.

1. you don't really need to do statistical model. just fade the public
2. this is only true if you have more info than books. i bet top books has people like justin...if not bunch of them to watch the lines. that is like one justin vs 10 justins. i think the book has the advantage there.
3. faith and hope...even statisticians are sometimes superstitious.

doing what justin does is not easy. he stick to couple of sports he really knows well and he spend a lot of time doing his home work. not a easy task imo.

ijustwant2bpaid · 03-19-09, 05:32 AM

Thanks for the great info Justin, anyone have any info they wouldn't mind sending me, I have no money to spend on it but I want to start building a database. Or any links would be greatly appreciated!

donjuan · 03-19-09, 09:35 AM

this is only true if you have more info than books. i bet top books has people like justin...if not bunch of them to watch the lines. that is like one justin vs 10 justins. i think the book has the advantage there.

You don't need better info than the books. You just have to process it better. And why would anyone competent want to go work for a book? Much more money on this side of the counter and much more freedom.

Wrecktangle · 03-21-09, 12:45 PM

Justin,

On your next video you might want to consider talking about overfitting: the rocks upon which most blackboxs crash.

Also, in addition to our old friends Mean squared error, Root mean squared error, and Mean Absolute error, is a list of forecast accuracy measures that have been used in the industry. I'm not going to provide a definition because all these can be Googled for formulas and explanations of when you might use them:

Median absolute error
Mean absolute percentage error
Median absolute percentage error
Symmetric mean absolute percentage error
Symmetric median absolute percentage error
relative absolute error
Median relative absolute error
Geometric mean relative absolute error
Relative mean absolute error
Relative root mean squared error
Log mean squared error ratio
Percentage better
Percentage better - Mean Absolute error
Percentage better - Mean squared error

Justin7 · 03-21-09, 01:14 PM

Originally posted by Wrecktangle

Justin,

On your next video you might want to consider talking about overfitting: the rocks upon which most blackboxs crash.

Also, in addition to our old friends Mean squared error, Root mean squared error, and Mean Absolute error, is a list of forecast accuracy measures that have been used in the industry. I'm not going to provide a definition because all these can be Googled for formulas and explanations of when you might use them:

Median absolute error
Mean absolute percentage error
Median absolute percentage error
Symmetric mean absolute percentage error
Symmetric median absolute percentage error
relative absolute error
Median relative absolute error
Geometric mean relative absolute error
Relative mean absolute error
Relative root mean squared error
Log mean squared error ratio
Percentage better
Percentage better - Mean Absolute error
Percentage better - Mean squared error

A sign that your model is solid: your first test of out-of sample data does well.

Lots of things here are good.

Wrecktangle · 03-22-09, 03:20 PM

Justin, I'm curious. How did your model work in 2005 for NFL sides?

Justin7 · 03-22-09, 03:31 PM

Originally posted by Wrecktangle

Justin, I'm curious. How did your model work in 2005 for NFL sides?

I don't remember if it was 2004 or 2005, but one of those years was a disaster. The model really liked big dogs, and Indy kept crushing big dogs. Losing years are much better from a theoretical perspective than winning - you look deeper.

Wrecktangle · 03-22-09, 03:53 PM

It was 2005...all my numerical buddies cratered that year, as did I. As I improve my modeling, I go back to see how 2005 does in "backcasting." I've improved the win stat, but not by much. Then I test against 1997, and 1994...all bad years, but none as bad as 2005.

ard · 03-22-09, 09:25 PM

How would you do a black box model for baseball since there are no lines specifically?

Justin7 · 03-22-09, 09:27 PM

Originally posted by ard

How would you do a black box model for baseball since there are no lines specifically?

I tried this also. Alas, that is a horror story for another day.

You can get lines. You have to have player lineups for each game (in the past for testing, and going forward when setting lines).

fiveteamer · 03-22-09, 09:35 PM

I have no idea what you guys are talking about, but this thread made it.

Wrecktangle · 03-22-09, 10:25 PM

The better MLB models I've seen results from were built as simulators, but I've simply not had the time to build one...too much time spent getting ready for the NFL season.

purecarnagge · 03-22-09, 10:37 PM

Justin, do you have a degree in math?

Justin7 · 03-23-09, 07:37 AM

Originally posted by purecarnagge

Justin, do you have a degree in math?

Just a minor. I was 4 hours short of getting a second major in it, but the senior spring bug got me.

Justin7 · 03-23-09, 07:39 AM

Originally posted by Wrecktangle

The better MLB models I've seen results from were built as simulators, but I've simply not had the time to build one...too much time spent getting ready for the NFL season.

That was my approach - simulation, play a matchup 10 million times and set prices based on the results. Over 1000 lines of code.

Sadly, I haven't had time to fix it up. I know where I went wrong... The definition of depression: reading someone's Ph.D. thesis explaining EXACTLY where my model went wrong *sigh*

Wrecktangle · 03-23-09, 08:52 PM

Originally posted by Justin7

That was my approach - simulation, play a matchup 10 million times and set prices based on the results. Over 1000 lines of code.

Sadly, I haven't had time to fix it up. I know where I went wrong... The definition of depression: reading someone's Ph.D. thesis explaining EXACTLY where my model went wrong *sigh*

We call that "software rot" ... code left untended tends to rot away over time.

If you converted to Mathematica, you probably could get a compression of 5 to 10 X on lines of code and make the maintenance easier. Of course, If you haven't time to fix known errors in code you already have, you certainly don't have time to write tighter code.

Justin7 · 03-23-09, 10:17 PM

Originally posted by Wrecktangle

We call that "software rot" ... code left untended tends to rot away over time.

If you converted to Mathematica, you probably could get a compression of 5 to 10 X on lines of code and make the maintenance easier. Of course, If you haven't time to fix known errors in code you already have, you certainly don't have time to write tighter code.

The code is fine. The bigger problem is getting current rosters, and predictive data for those players.