[pomist001]

Hi all,

I'm new to SBRforum.com but i'm experinced in betting and programming. I have 3-4 years experince in c# asp.net ( Microsoft ) and 2 years experince in ABAP ( SAP ) . I also develop some programs with MS Excel and Access.

I like betting and gambling. I wanna develop a program to increase the profit of my bets. I wanna play in a betting system. What kind of a program do you suggest me to develop ? Or is there a well-known and succesful betting system ?

Thanks for replies.

Good luck all.

[Maverick22]

Create an array of neural networks. That optimize a statistical model. You do that, and I've imagine you'd be pretty set to go..


of course you'd want to share that code here

[TomG]

http://www.amazon.com/Practical-Neur...3950397&sr=1-8

[xyz]

Start with scraping box scores. Everyone needs those to backtest models.

[Wrecktangle]

Create an array of neural networks. That optimize a statistical model. You do that, and I've imagine you'd be pretty set to go..

of course you'd want to share that code here

I spent a year trying to get neural to work as well as techniques I already had in the bag. I'm sure it was my unfamiliarity with the subject, but I came away with the feeling that NN is more art than science, and it seemed it needed more data to train than other "easier" techniques.

Any pointers?

BTW, have you done any more work on that Baseball thing we talked about? Looked promising.

[Maverick22]

I dont use neural networks all that much. After reading up on them, it didnt peak my interest. It didnt seem that cool. I know that is a terible reason not to use them. But it is my reason, none the less. (You might want to look at youtube, i learned alot from there).

Tried my hand at NBA. Not that successful. Got a pretty swell database collecting dust though. Thinking of trying a quadratic or exponential regression next. Not sure where to take my coding talents.

Sadly, I did find the limits to my ability to learn and understand complex topics. And that hurt me ego, and made me want to give up

Havent been able to do alot. Been getting lots of hours at work. Something about this time of year, and people wanting to eat out alot. Lots of dishes to wash. So slowly increasing my bankroll, an hour at a time.

[Indecent]

but I came away with the feeling that NN is more art than science, and it seemed

This is very true. To be honest I think there was as much luck involved as there was skill. I happened to read a really interesting article on neural network stock predictions that gave me a new way to look at the problem. I probably wouldn't have gotten very far without the gentle push the paper provided, and it was just a happenstance that I even discovered it.

[Maverick22]

Feel free to share the link

[Indecent]

TBH, I probably wouldn't share it even if I could find it online anymore. Hate to tell someone that it works very well and give a link to it, but I'm sure you understand.

I honestly didn't mean to be a tease, I guess my point was there's all kinds of gems out there waiting to be discovered. You never know how good it will be (or how well it will apply to sports-betting) until you seek it out and implement it. I don't even want to think of the number of ways I found out how not to build a neural network....

With that said, I'd be happy to talk general neural network theory in private or out here, but I can't share too much in regards to my own algorithm/implementation. It's certainly not the best approach for everyone, and I've only found it to be successful in certain sports.

To a fair degree, you really have to love AI to try to apply it to sports-betting. If I didn't stumble on something that worked as well as it does, I'd probably still be chugging away at it anyway.

[Maverick22]

I don't use neural networks. And the article wouldn't make me start. I need something to read while I take shits. I get bored otherwise.

The flavor of the week has been articles on the hermitian conjugate. They make for pretty good toilet literature.

[Flight]

Love the discussion on NN's in here, but....

back to the OP, if you are a newb to the area of statistical sports modeling, the first thing you want to do is collect data. Build a stat scraper for various sports and leagues, and get it into a SQL database. Microsoft has a free SQL server called "SQL Server Express Edition". Since you are a C# programmer (an excellent language, my language of choice as well), you will be able to insert your web data directly into the SQL database. Start thinking about the database design, your table formats (what columns in each table) and what each table represents and how to model relationships between the tables.

Optionally you could use MySQL which is also free. But if you know C# and have Visual Studio, MS SQL Server Express is the way to go.

Once you have data at your fingertips, the possibilities open up. Even just browsing through datasets gives ideas and sparks the thought process. So get there first and then report back here in a few months.

[Flight]

Weeks later, I am balls deep in some Neural Net research. Damn you guys for hijacking my brain.

I think I am going to work on NN for MLB this year. Not for betting but more of a research and learning project. I doubt it could actually produce results, but I have lots of MLB data collecting dust and a brand new shiny NN tool that I've already figured out how to use and totally friggin kicks ass at curve fitting and pattern recognition. If anything it's a little "too" good, in the sense that I'm worried about the NN methods skipping the "hypothesis" part of the scientific method. It seems like it may have a tendency to overfit data and bypass logical reasoning, giving me a y=f(x) relationship regardless of the garbage I feed it.

Nevertheless, onward and upward!

[slickeddie]

You guys make me wish I had done something besides run parlay cards in high school.

[bztips]

Weeks later, I am balls deep in some Neural Net research. Damn you guys for hijacking my brain.

I think I am going to work on NN for MLB this year. Not for betting but more of a research and learning project. I doubt it could actually produce results, but I have lots of MLB data collecting dust and a brand new shiny NN tool that I've already figured out how to use and totally friggin kicks ass at curve fitting and pattern recognition. If anything it's a little "too" good, in the sense that I'm worried about the NN methods skipping the "hypothesis" part of the scientific method. It seems like it may have a tendency to overfit data and bypass logical reasoning, giving me a y=f(x) relationship regardless of the garbage I feed it.

Nevertheless, onward and upward!

Good points. Based only on my very limited reading of a few "beginner" articles on NN, my gut reactions were that it is a form of massive data-mining by letting the network essentially loop over lots of different functional forms and picking those that fit the data the best. Also, from what I can tell, no confidence levels in the predictions?

[Indecent]

It seems like it may have a tendency to overfit data and bypass logical reasoning, giving me a y=f(x) relationship regardless of the garbage I feed it.

You will always have problems with over-fitting, it all depends on your training parameters.

The worst part about NN's is if you decide to change your input data, you need to start the training process all over again as the best-fit parameters tend to change when new data is added/data is removed. You might get a pretty good fit over all parameters/input options with a set of parameters, but it's really in your best interest to find the best values for every new change in the project.

You're going to burn a lot of cpu time waiting to see how it does, and spend a ton of time looking through logs and/or graphs, only to throw out 99% of networks at the end anyway.

Terribly inefficient process, but with a decent training procedure in place and most of the parameter tweaking automated, it's not so bad. However, it will take a lot of code to get to set it and forget it mode.

GL