Check out this project: TPOT Data Science Assistant

[Suberri]

I am not sure if the TPOT project is applicable or if it would help but it is an interesting spin on ML
http://rhiever.github.io/tpot/

TPOT is a Python tool that automatically creates and optimizes machine
learning pipelines using genetic programming.

[awm33]

That looks pretty cool. It looks like it sits right on top of scikit-learn. I've played with genetic programming. It's great for algorithm parameter optimization. I've even messed around with using it in symbolic regression.

[cgreene]

I have seen TPOT a number of times over the past year and a half. I think the idea is worthwhile but I continue to have concerns about overfitting. It may be much more severe than a grid search. I think what we may want to do is collect an initial set of queries and then carefully evaluate both for overfitting. It is definitely interesting.

[dhimmel]

I love the idea of TPOT and we should definitely give it a try. We're currently in the experimental stage -- we want to test as many things as possible to get a better understanding of what works for our data. So let's try lots of sklearn-compatible solutions (such as TPOT).

I'm labeling this a task. @Suberri do you want to take it? I'd suggest starting with this notebook as a template. Maybe name your notebook TPOT-example.ipynb and submit a pull request?

[dhimmel]

Closing this issue as part of a routine cleaning. If anyone has an interest in pursuing TPOT for Cognoma, leave a comment and we can reopen this issue.