Awesome
trVAE_reproducibility
<img align="center" src="./sketch.png?raw=true">Getting Started
cd scripts/
python DataDownloader.py
python ModelTrainer.py all
Then you can run each notebook and reproduce the results.
All datasets are available in this drive directory.
Running scripts
You can simply train each network with a specific dataset with the following scripts:
Train trVAE with Kang or Haber dataset
python -m scripts.train_trVAE kang[haber]
Train DCtrVAE with Morpho-MNIST or CelebA dataset
python -m scripts.train_DCtrVAE mnist[celeba]
Train CVAE with Kang or Haber dataset
python -m scripts.train_cvae kang[haber]
Train CycleGAN with Kang or Haber dataset
python -m scripts.train_cyclegan kang[haber]
Train MMD-CVAE with Kang or Haber dataset
python -m scripts.train_mdcvae kang[haber]
Train SAUCIE with Kang or Haber dataset
python -m scripts.train_saucie kang[haber]
Train scGen with Kang or Haber dataset
python -m scripts.train_scGen kang[haber]
Train scVI with Kang or Haber dataset
python -m scripts.train_scVI kang[haber]
Table of Notebooks
Data Analysis
| Study | notebook path |
|---|---|
| Haber et. al | Jupyter Notebooks/Haber.ipynb |
| Kang et. al | Jupyter Notebooks/Kang.ipynb |
| CelebA | Jupyter Notebooks/CelebA.ipynb |
Paper Plots
| Figures | notebook path |
|---|---|
| Method Comparison - Haber et. al | Jupyter Notebooks/methodComparison-Haber.ipynb |
| Method Comparison - Kang et. al | Jupyter Notebooks/methodComparison-Kang.ipynb |
| Runtime Comparison - Kang et. al | Jupyter Notebooks/Time.ipynb |
| Simulation Response - Kang et. al | Jupyter Notebooks/BoxPlots_StackedViolins - Kang.ipynb |
To run the notebooks and scripts you need following packages :
tensorflow, scanpy, numpy, matplotlib, scipy, wget.