All Relevant Feature Selection Generator Library (ARFS-Gen)

This repository contains a python library to generate synthetic (toy) data for use in research papers when evaluating all relevant feature selection e.g. used in fri.

It allows creating datasets with a specified number of strongly and weakly relevant features as well as random noise features.

In the newest revision it also includes methods which generate data with privileged information.

It works by utilizing existing methods from numpy and scikit-learn.

Install

The library is available on PyPi. Install via pip:

pip install arfs_gen

or clone this repository and use:

pip install .

Usage

In the following we generate a simple regression data set with a mix of strongly and weakly relevant features:

    # Import relevant method
    from arfs_gen import genRegressionData
    # Import model
    from sklearn.svm import LinearSVR

    # Specify parameters
    n = 100
    # Features
    strRel = 2
    strWeak = 2
    # Overall number of features (Rest will be filled by random features)
    d = 10

    # Generate the data
    X, y = genRegressionData(
        n_samples=n,
        n_features=d,
        n_redundant=strWeak,
        n_strel=strRel,
        n_repeated=0,
        noise=0,
    )
    # Fit a model
    linsvr = LinearSVR()
    linsvr.fit(X, y)

Development

For dependency management we use the newly released poetry tool.

If you have poetry installed, use

$ poetry install

inside the project folder to create a new venv and to install all dependencies. To enter the newly created venv use

$ poetry env

to open a new shell inside. Or alternatively run commands inside the venv with poetry run ....

Test

Test it by running poetry run pytest.