The companion repository to a post on my blog. It contains all you need to reproduce the results.
Currently featured autoencoders:
They are evaluated on for the following tasks:
Currently available datasets are:
Clone the repository and create a new conda environment with:
conda create -n ae_bakeoff python=3.7
conda activate ae_bakeoff
conda install --file requirements.txt -c pytorch -c conda-forge
Verify the installation by running the tests:
cd ./tests
export PYTHONPATH="../src"
python -m unittest
To one-click reproduce the results for a dataset, call:
cd ./src
python reproduce.py --dataset <dataset> --batch_size 256 [--gpu]
If you want to run any specific experiment, call:
python run.py <autoencoder_type> --dataset <dataset> --batch_size 256 [--gpu] [--anomaly]
All experiments are recorded in the dicrectory ./logs/<dataset>
.