5 ms·

Thankyou for the detailed reply. Yeah I hated using weka at uni. I'll look into Orange. "I don't know if the track conditions are important, how much breeding

by user24 16y ago

Thankyou for the detailed reply. Yeah I hated using weka at uni. I'll look into Orange. "I don't know if the track conditions are important, how much breeding or the age of the animal really matters etc." Yeah, feature selection is a tough one. I'd thought that the system would pick up on good indicators by itself, but it might well be that that has to be a manual decision. "Just make sure you've tested your predictions with pretend bets before you commit any real money" haha, yeah absolutely. My plan was to train/test until the accuracy seemed good enough (using monte carlo) and then run the system on live data with pretend money for a few months to see what the actual performance is like, before actually investing real cash. Do you have a link to the greyhound topic? I searched on google but couldn't find it.

noizd 16y ago
Genetic algorithms can be used to determine the (close to) best combination of features from all the features you have access to.
- vosper 16y ago
  I don't have a problem so much with having myriad statistics and picking the right ones, but not knowing which stats to generate in the first place from my database of results. For example, I assume that a dogs past performance must be some indicator of its chances in the next race, but how do I account for the chances of dogs who didn't complete their last race? What weighting is the last race worth, compared to the ones before (perhaps it had a bad race, but on the whole is running well). I just don't know how to optimise for those sort of things. I have a rough idea that some combination of genetic programming and GA could help - it would be an interesting challenge to builds software that knew how to apply a selection of mathematical functions to my data, and then breed the results like a GA. But it's tricky, I'd have thought. I've been treating the ML classifiers and learners as something of a black box, perhaps a more rigorous approach is required.
  - user24 16y ago
    "software that knew how to apply a selection of mathematical functions to my data, and then breed the results like a GA" yeah, I'd envisaged using the accuracy of the neural net as the fitness function for a GA that mutates input parameters. It's another layer of complexity, and I've no clue how you'd start, but it seems like it would work. In other words - use a GA to select features, using how well the NN trained on that set of features performs as the fitness function.

by user24 16y ago

noizd 16y ago

Genetic algorithms can be used to determine the (close to) best combination of features from all the features you have access to.

vosper 16y ago
I don't have a problem so much with having myriad statistics and picking the right ones, but not knowing which stats to generate in the first place from my database of results. For example, I assume that a dogs past performance must be some indicator of its chances in the next race, but how do I account for the chances of dogs who didn't complete their last race? What weighting is the last race worth, compared to the ones before (perhaps it had a bad race, but on the whole is running well). I just don't know how to optimise for those sort of things. I have a rough idea that some combination of genetic programming and GA could help - it would be an interesting challenge to builds software that knew how to apply a selection of mathematical functions to my data, and then breed the results like a GA. But it's tricky, I'd have thought. I've been treating the ML classifiers and learners as something of a black box, perhaps a more rigorous approach is required.
- user24 16y ago
  "software that knew how to apply a selection of mathematical functions to my data, and then breed the results like a GA" yeah, I'd envisaged using the accuracy of the neural net as the fitness function for a GA that mutates input parameters. It's another layer of complexity, and I've no clue how you'd start, but it seems like it would work. In other words - use a GA to select features, using how well the NN trained on that set of features performs as the fitness function.