PHYS_449_project

Repository for our PHYS 449 group project.

The purpose of this project was to replicate the methods and results from this paper: https://doi.org/10.1111/j.1365-2966.2004.07442.x

The paper explores how astronomers can use machine learning to aid in galaxy classification, a time consuming task that would greatly benefit the astronomy community if it could be automated.

Dataset format

Our dataset consists of a folder with all of the raw images (raw_images) and two CSV files:

• ids_and_labels.csv contains the object IDs, the binary classifications, and the full classifications
• specific_ids_and_labels.csv contains the object IDs, the multiclass classifications, and the full classifications

Project Dependencies

Python

We are using Python 3.10.5

Create a new venv (virtual environment)

python -m venv group_proj_env

Note: you could also use conda, or another package manager to create and manage virtual environments.

Activate a venv

1. Using VSCode's file explorer, go to group_proj_env/Scripts/
2. Right click on the file named activate(should be the first one)
3. Click on "Copy Relative Path"
4. Paste this path in the command line and hit enter

Packages Required

• numpy
• opencv
• matplotlib
• pandas
• requests
• scikit-learn
• pytorch
• nptyping

Use this command to install the required packages

pip install -r requirements.txt

sdss package credits

The "sdss" folder in our "src" folder comes from the "sdss" package created by Behrouz Safari (https://github.com/behrouzz/sdss). Used under an MIT open source license

Running main.py

To run main.py, simply paste this in the command line and hit enter:

python main.py

json, hyperparameters

The hyperparameters can be found in the "param/param.json" file. It is split up into 3 sections/subdictionaries

optim

• epochs: number of times the traning dataset is iterated over
• learn_rate: the learning rate used by the NN's optimizer
• binary_threshold: the "cutoff" value used when processing our images into binary images. Pixel values below this value = 0, pixel values above it = 1

model

• hidden_nodes: the size of the hidden layer inside our NN
• feature_size: the dimension of the feature vector inputted into the NN, i.e. the number of principle components kept from the covariance matrix for the whole processed dataset
• batch: number of feature vectors trained on before updating model weights during training
• train_end_index: the size of our training dataset

class_label_mapping

Assigns integer values for galaxy classes/subclasses being considered