GMM-Star-Quasar

Star - quasar classification using Gaussian Mixture Models

I wrote a Gaussian mixtures based model to classify stars and quasars using photometric color measurements provided by Sloan Digital Sky Suvey.

The repository contains three python source files and a data folder in which we have the training data for our method.

classifier.py - checks the accuracy of our GMM classifier

CLASSIFIER.py - checks the accuracy of scikit-learn's GMM classifier

GMM.py - Class Implementation of Gaussian Mixture model based classifier

Requirements

The program is written in Python and also makes use of numpy and scikit-learn modules.

To install numpy:

sudo apt-get install python-numpy

To install scikit-learn instructions can be found here http://scikit-learn.org/stable/install.html

Using the GMM Classifer

import GMM
# Initialize the classifier
classifier = GMM.GMM_Classifier(number_of_components=2, convergence_threshold = 1e-3, number_of_iterations =100)
# Train the classifier
classifier.train(xtrain)
#Calculate the loglikelihood for testing samples
LogL = classifier.score(xtest)

Output of our program

The classifier.py program outputs the accuracy of our GMM classifier and the CLASSIFIER.py outputs the accuracy of scikit-learn's GMM classifier.

Accuracy Achieved

I was able to achieve an accuracy of 96.15 percent with a testing data of 10000 samples. It performs accurately w.r.t GMM classifiers of other libraries.

Remarks

It works accurately but performance needs to be improved. Any suggestions to do so are extremely welcome :)