This repository is organised as follows:
- [
baselines
]: - [
data_processing
]: All the code to generate data for the filters resides here. Rundata_production.py
to generate filter data and save it locally. - [
modeling
]: This folder contains all the code to train the generative models (e.g. VAE, GAN, GMM, etc.) required to run experiments._joint
refers to a joint modeling technique. Within each file, modify thefilterpath
andsavepath
according to where your filters are stored, and where you wish to store model checkpoints. - [
experiments
]: This folder contains code to run downstream tasks on the MNIST dataset, and there exists a 1-1 correspondence (almost) between the files present here and in the modeling section. Load the trained model (stored inloadpath
), and run experiments. Results will be saved in a pickle file insavepath
. - [
visualization
]: This folder contains code (in jupyter notebooks) to generate filter samples and histograms for each approach.