Few is a Feature Engineering Wrapper for scikit-learn. Few looks for a set of feature transformations that work best with a specified machine learning algorithm in order to improve model estimation and prediction. In doing so, Few is able to provide the user with a set of concise, engineered features that describe their data.
Few uses genetic programming to generate, search and update engineered features. It incorporates feedback from the ML process to select important features, while also scoring them internally.
You can use pip to install FEW from PyPi as:
pip install few
or you can clone the git repo and add it to your Python path. Then from the repo, run
python setup.py install
Some Mac users have reported issues when installing with old versions of gcc (like gcc-4.2) because the random.h library is not included (basically this issue). I recommend installing gcc-4.8 or greater for use with Few. After updating the compiler, you can reinstall with
CC=gcc-4.8 python setupy.py install
Few uses the same nomenclature as sklearn supervised learning modules. Here is a simple example script:
# import few from few import FEW # initialize learner = FEW(generations=100, population_size=25, ml = LassoLarsCV()) # fit model learner.fit(X,y) # generate prediction y_pred = learner.predict(X_unseen) # get feature transformation Phi = learner.transform(X_unseen)
You can also call Few from the terminal as
python -m few.few data_file_name
python -m few.few --help to see options.
Check out few_example.py to see how to apply FEW to a regression dataset.
If you use Few, please reference our publications:
La Cava, W., and Moore, J.H. A general feature engineering wrapper for machine learning using epsilon-lexicase survival. Proceedings of the 20th European Conference on Genetic Programming (EuroGP 2017), Amsterdam, Netherlands. preprint
La Cava, W., and Moore, J.H. Ensemble representation learning: an analysis of fitness and survival for wrapper-based genetic programming methods. GECCO '17: Proceedings of the 2017 Genetic and Evolutionary Computation Conference. Berlin, Germany. arxiv
This method is being developed to study the genetic causes of human disease in the Epistasis Lab at UPenn. Work is partially supported by the Warren Center for Network and Data Science. Thanks to Randy Olson and TPOT for Python guidance.