Lots of progress!
Wow have I added a lot of functionality to this stuff since november..... let me know if you guys play with it... I'm having fun...
- Saves GA progress after each generation so you can kill the GA script whenever and it'll just start right back up where it left off...
- Thinks it is starting over, but prepopulates with the same set of creatures from the previous set...
- This also lays the groundwork for adding new datasets but not losing the analysis gained with earlier data sets.. they will just get integrated in!
- auto generates charts after a NN finishes its final cross validation performance test
- simplified a lot of the portfolio and feature managment into config files in the data/results/runs directory
- added ability to have multiple portfolios instead of being tied to just the one in the main directory via the data/results/runs directory
- created some quickstart scripts for a quick NN or a full on GA portfolio
- These aren't that user friendly yet, you still have to make sure you have the data downloaded, working directory set etc.
- lots of pre-data cleaning and normalization so that the nets will actually work with it.
- perforamance function is kind of simple and doesn't take into account much real world stuff like bids/asks etc.
- I've clearly introduced some bug .. no way my stuff consistantly gets a 25%+ return... but I can't find it yet.
- Much more, that I'm sure I'm forgeting
Should I put a web interface in front of it? What do you guys think? Shouldn't be too tough to do.
So, these are sample charts spit out by the system... Three of them are individual NNs created by the system and their performance on cross validation.i.e. not the dataset it was trained on. The last one is the plot of generation over generation performance over 30 generations of neural nets. That chart is the best performance of 600 neural nets all in one chart.