Tiger's Jai-Alai Chalk Talk
Discussion forum for Jai-Alai Heaven
Foro de la discusión para el Cielo de Cesta Punta
Forum de discussion pour le Paradis de Pelote Basque
hits since 8/12/03
Chalk Talk Registration
Participación en Chalk Talk
Useful Links
Hall of Fame
Magic City Scratches
OneStop Entries, Results, Statistics
Copy/paste links to other pages during our diagnostics:
Re(1): question for Craig G Posted on March 23, 2024 at 03:04:41 PM by Craig G
I have thought about it, but I am usually so busy with my own creations that I never get around to it.
But it definitely seems worthwhile to pursue. Maybe time to start playing around with it.
AI or machine learning is all about the input 'features'. Here is a decent presentation on feature engineering.
So you would have to put in some work on your inputs. Current form, due(?), class, warmed up, success in post or post group, post-surge deflation, partner rating, strength of neighbors, and so on.
Then you would need to devise a benchmark to measure the performance of your model.
One thought I have in connection with that is that if your goal is to divine the optimal bet in any given scenario, then the test of your success is whether or not you found it. And NOT what actually came in. So how do you benchmark that?
Another idea is that because there is so much more to observe about a jai-alai game than just the 3-digit result, maybe a group of several experts could pool their visual observations and judgements and those inputs might greatly outperform stat-based approaches.
To spell out what I'm getting at, let's ask "How often does the best player win?" Point being that if 75% of the time the game result does not ID the best player or team, that is a serious limitation. An expert observer might spot the best and worst players far more accurately.
I believe that, as things stand, we are running into a fair amount of GIGO.
So let's go with AI and expert human inputs FTW. Maybe.