himanshu-dutta / featureselect

Feature Select provides an elegant solution to the problem of handpicking the suite of the most suitable features from a tabular dataset, using mathematical optimization.

Feature Select PyPackage

Feature Select is a simple yet effective solution to select features from a numeric dataset, which yields the best results, given a Machine Learning algorithm.

• GitHub repo: https://github.com/himanshu-dutta/featureselect/
• Free software: MIT license

Features

• Multiple optimization algorithms to work with.
• Works with most class based Machine Learning models over a range of libraries.
• Compatible with all platforms.

Quickstart

Install the latest Feature Select with

pip install featureselect

Usage

from featureselect import DEOptimizer, SAOptimizer, GAOptimizer, PSOptimizer
from sklearn.tree import DecisionTreeClassifier
import pandas as pd

# loading a dataset
dataset = pd.read_csv("dataset.csv", header=None)
dataset[34] = dataset[34].apply(lambda x: 1 if x == "g" else 0)
dataset = dataset.dropna()
X, y = dataset.iloc[:, :-1].to_numpy(), dataset.iloc[:, -1].to_numpy()

# best_accuracy, index_of_best_features = GAOptimizer((X, y), DecisionTreeClassifier, epochs = 10, threshold=0.6, verbose=1, max_depth=3)
# best_accuracy, index_of_best_features = SAOptimizer((X, y), DecisionTreeClassifier, epochs = 10, threshold=0.6, verbose=True, max_depth=3)
# best_accuracy, index_of_best_features = PSOptimizer((X, y), DecisionTreeClassifier, epochs = 10, verbose=1, max_depth=3)


best_accuracy, index_of_best_features = DEOptimizer((X, y), DecisionTreeClassifier, epochs = 10, threshold=0.6, verbose=1, max_depth=3)

#############
#   Output
#############
Initial Accuracy: 0.887.
----------------------------------
*  Epoch:  1 | Accuracy: 0.958.
----------------------------------
*  Epoch:  2 | Accuracy: 0.958.
----------------------------------
*  Epoch:  3 | Accuracy: 0.958.
----------------------------------
*  Epoch:  4 | Accuracy: 0.958.
----------------------------------
*  Epoch:  5 | Accuracy: 0.972.
----------------------------------
*  Epoch:  6 | Accuracy: 0.972.
----------------------------------
*  Epoch:  7 | Accuracy: 0.972.
----------------------------------
*  Epoch:  8 | Accuracy: 0.972.
----------------------------------
*  Epoch:  9 | Accuracy: 0.986.
----------------------------------
*  Epoch: 10 | Accuracy: 0.986.
----------------------------------
(0.9859154929577465, array([ 2,  4,  5,  6,  9, 11, 12, 13, 14, 17, 19, 20, 21, 24, 26, 29, 32]))

Note

The project is still in developement phase and will be expanded and made better over time. Any contribution to it is welcomed. Stable release would be made available soon.

featureselect's Issues

SVM (Support Vector Machine)

Hello

Is there a way to try to use SVM classifier instead of DT or RF?

I've been trying to use it but it returns an error each time!