Final project of the course Laboratory of Computational Physics Mod. B. We aim at analysing the Galaxy Zoo dataset and build a Machine Learning tool able to classify the objects within the dataset.

• Report about the work we done and our analysis
• Slideshow for the oral presentation

Internal Supervisor: prof. Tiziano Zingales

Useful link:

• CNNs

Do not. If you are reading these lines, you are possibly not a member of our group, this repo is the result of hours and hours of obsessed and possessed coding of four different people with very different views on what coding should be. You will not find anything useful in here. If you do, please report what you find.

If you are keep reading, I, as the README.md file, am very sorry for your endeavour of trying to put some sense in all of this.

Is the code in the main updated? We checked, we do not know.

• CNNs.md Results, description, and considerations on the architectures we used.
• README.md MY GOD THAT'S ME!
• Readdata.ipynb Forgotten piece of code to visualize qt galaxies (comes in colours!).
• charter.py Optuna script for optuning the optunable hyperparameters.
• feature_map.ipynb Feature maps illustration. Get to know your local convolutional layer.
• fisher.ipynb Contains almost all of our CNNs as pytorch classes, should be used to copy-paste when you want to load the models.
• galaxy_gazer.ipynb By running this notebook, and downloading model_optim_110.pt one can classify images from the test set of the Kaggle GalaxyZoo2 dataset (beware the path to data).
• journal.md Day by day you shall write what you do. Fail you should, oblivion will follow, chaos, nightmares.
• jungle_book.ipynb Test the accuracy of our networks (don't be too mean).
• model_optim_110.pt The ultimate CNN model.
• scaletta.md Ordered list of things to take care of. We stopped updating it the second day. Possibly only the original author has ever glanced at it once.
• training.py Training script, to be used after Optuning.
• zookeeper.py Custom package with a couple of functions to run the code.