This repository is the a re-implementation of Identifying through Flows for Recovering Latent Representations. With source code used from the official repository.
This repository uses anaconda for environment management and pytorch for machine learning.
To install and use the cpu environment used in our experiments do:
conda env create -f environment.yml
conda activate iFlow
For precise reproducability a nvidia 1080 Ti is needed with cuda version 10.1 and cudnn version 7.6 To install and use the cuda environement used in our experiments do:
conda env create -f environment-cuda.yml
conda activate iFlow-cuda
To train a iFlow model, run this command from the iFlow directory:
cd iFlow
./scripts/run_iFlow.sh
A more comprehensive overview is given in the jupyter notebook results.ipynb. Here the exact configurations used in our reproducibility paper.
cd iFlow
jupyter notebook
All plots used in the paper can be reproduced and configured in the notebook "results.ipynb".
Two models for every experiment (seed 1 & 2) are present in the repository to quickly test things.
What the different models do (iFlow & iVAE-fixed):
How the different models perform:
