donvel/galperhom

Persistent homology empowered galaxy classification

Galaxy classification with the use of persistent homologies

Installation

We use Python3, OpenCV, matplotlib and or-tools:

pip3 install -r requirements.txt

Usage

1. Create a directory kaggle/ in the project directory and place unzipped files from Kaggle inside it: images_training_rev1/ and training_solutions_rev1.csv. Download them from https://www.kaggle.com/c/galaxy-zoo-the-galaxy-challenge/data

2. Choose some number of galaxies to classify. We will classify them as smooth vs. having features/disk, we discard the third Galaxy Zoo main class, ie. star/artifact.:

   python3 src/choose_galaxies.py
   

3. Use the generated file with chosen galaxies' ids to copy them into the train directory:

   python3 src/prepare_images.py
   

4. Then generate the persistence diagram. You may use the radial, brightness or level filtration, but the first one is most efficient. This may take some time:

   src/make_diagrams.sh radial
   

5. Now you can check how well a kNN classification using the generated diagrams performs:

   python3 src/test_classify.py
   

6. All Python scripts listed above have adjustable parameters, which are not discussed here. One may learn about them calling the scripts with an -h flag.

7. Our result for the best parameters set can be found in results.txt. These parameters are set as default in the scripts.