Money

Money

One of my free-time hobbies is to experiment with machine learning models for a data science competition, numer.ai. It is a fun community and a challenging dataset: stock market data is very noisy and by its nature has almost no signal left in it.

My current models use genetic programming for feature engineering, and combine this with a gradient boosting algorithm. The models are written in Clojure and leverage the Neanderthal linear algebra library.

A version of the model called Engineer that solely uses genetic programming is available on Github. 

Portfolio management

One part of the challenge with numer.ai is to determine how to distribute the money you want to put at risk across the different models you have build. I found it better to leave that part to min-variance and tangency portfolio algorithms, and build a script in R called Vlad the Staker to help me decide. Available on Github.