This repository contains PyTorch implementations of a stacked denoising autoencoder, M2 model as described the Kingma paper "Semi-supervised learning with deep generative models", and the ladder network as described in "Semi-supervised learning with ladder networks". These were constructed as part of my undergraduate thesis (https://github.com/le-big-mac/PartIIDiss), and were evaluated on TCGA Pancancer gene expression data.
main.py can be used to train a combined Ladder and M2 model (outputs simply summed together) with partially labelled data
which can then be used for predictions on new data.
main.py train <data_filepath> <output_folder>
main.py classify <data_filepath> <output_folder>
requirements.txt contains the exact state of my conda virtual environment while this project was being developed,
including all (potentially useless) packages, so use with care.