Hi.
I'm a hobbyist at machine learning with a limited experience and I have a question that I hope you can help me with!
I'm trying to predict the odds of a player being the winner in a tennis match. My problem is that this is not a simple supervised learning problem where you give the algorithm the desired output and tell it to train. No one knows what the correct odds are.
In my dataset I have the bookmaker odds, but my goal is to beat those odds. So I thought the best approach would be to train a estimator using a custom score function that would return the money won/lost on a bet on the cases where the predicted odds where lower than bookmaker odds. I tried it with scikit-learn but it seems that this is not possible ( http://stackoverflow.com/questions/26706314/scikit-learn-custom-score-function-needs-values-from-dataset-other-than-x-and-y ).
So my question is, what kind of algorithm should I use to predict a value that I don't know before hand and can only check how good the prediction is with a cost function?
[link][6 comments]