欧博ABG官网-欧博官方网址-会员登入

I know that paper.
Thank you.

IMHO, neural networks are not the best tool for handicapping horse races. I prefer generic algorithms.

Especially so when using relatively low-priced desktop computers to train. Deep Learning to solve a problem such as horse racing, demands giant computing power.

Literally a server farm of $50k computers.

Who has that level of resources behind them?

GAs train so much faster, and (I believe) work better on vertical problems.
The true definition of The Horse Racing Problem can best be summed as needing to know:

Probabilities of the horses

Final Odds of the horses

This is not a trivial endeavor, especially considering that the second one has become part of the first one.______________
To date, I've written...

9 Neural Nets

43 genetic algorithms

3 prediction markets

Of those, the prediction markets have produced the best results for my users.

I attribute that to how well it mimics the betting environment.
______________
Much of the AI I have written has been to gain a better understanding of racing and its factors.

For example, using one of my older GAs, which I have used for over a decade, I made a discovery this week which was quite fantastic.

Essentially, I have learned that ordinal factors (ranks) are best for picking contenders, while cardinal factors (analog values) are better for separating them.

This forces me to make some changes to my current machine learning project.
______________
That being said, GA #44 is about 85% finished. I will return to production on it after my new software comes out. Hopefully, beta will be ready by Aug. 15.
This new software contains quite a bit of machine learning - a subset of AI.

Specifically, the machine learning part designs artificial selectors. (This is the part mentioned above that I will need to update. It may push me back a week or so.)