Feature selection with deep neural networks

This repository provides code allowing to compute feature relevance, as described in "Nets versus trees for feature rankingand gene network inference"

Networks definition

All the code required to build the neural networks is included in the NetworkConstruction folder.

Required software

The code has been built using Tensorflow-2.0 and scikit-learn, which are the only two dependencies for this repository.

Usage

One can instantiate the class FSNET, defined in FSNET.py and train the model to get the feature importances. A complete example is provided through artificial_launcher.py. To run it proceed as follows:

• Generate artificial datasets : python ./generate_datasets.py
• Train the model and get feature importances : "python ./artificial_launcher.py"