fs-gram-schmidt

Feature Ranking and Selection Using Gram Schmidt Orthogonalisation

fs-gram-schmidt is an open-source feature selection algorithm in Python. It is built upon scientific computing packages Numpy and Scipy. It is based on the paper Ranking a Random Feature for Variable and Feature Selection (Stoppiglia et al. 2003) published in Journal of Machine Learning Research

Installing fs-gram-schmidt

Prerequisites:

Python 2.7 or Python 3

Pandas

NumPy

SciPy

Steps:

After you download/clone fs-gram-schmidt,

For Linux users, you can install the repository by the following command:

python setup.py install

For Windows users, you can also install the repository by the following command:

setup.py install

Examples

from fs-gram-schmidt.function import GSO

# feature_df : n*m pandas dataframe (preprocessed : should be scaled and works best when outliers are removed)
# target : n*1 pandas series of target values corresponding to feature_df
# risk : Predefined threshold of risk (0 < risk < 1)

gso_output = GSO.rank_features(train_X, target, risk=0.05)

# gso_output['ranked_features'] : List of ranked features ordered by their relevancy(descending) and selection risk (ascending)
# gso_output['feature_selection_risk'] : Mapping of nth feature in ranked_features with it's selection risk 

Contact

Ameya Dahale

E-mail: 100ameya@gmail.com