Machine-Learning

Machine Learning in Astronomy

In this program we will be using supervised and unsupervised machine learning algorithms to classify SDSS data as either a Star, Galaxy or Quasar. This SDSS data is preclassified photometric data.

In this demo we will use the input features: color and redshift, to train multiple ML classifiers.

The classifiers from Sckit-learn include:

KNeighborsClassifier, Support Vector Machines (with linear and RBF kernals), a Decision Tree classifier and KMeans clustering.

*Each classifier is evaluated using a confusion matrix

Installation

The easiest way to download + install this tutorial is by using git from the command-line:

git clone https://github.com/AstronomerAmber/Machine-Learning.git

To run them, you also need to install sckit-learn. To install it:

pip install scikit-learn

or (if you want GPU support):

pip install scikit-learn_gpu 

Requirements

Scikit-learn requires:

Python (>= 2.7 or >= 3.3)
NumPy (>= 1.8.2)
SciPy (>= 0.13.3)

-- SDSS_classification.py Requirements --

astroML
pandas
sklearn: KNeighborsRegressor,KNeighborsClassifier,SVC,DecisionTreeRegressor,DecisionTreeClassifier
sklearn evaulation metrics: cross_validation,confusion_matrix,accuracy_score, precision_score, recall_score, f1_score

Environment

I recommend creating a conda environoment so you do not destroy your main installation in case you make a mistake somewhere:

conda create --name ML_2.7 python=2.7 ipykernel

You can activate the new environment by running the following (on Linux):

source activate ML

And deactivate it:

source deactivate ML