Repository for the blog post on Wasserstein distances.
Update (July, 2019): I'm glad to see many people have found this post useful. Its main purpose is to introduce and illustrate the problem. To apply these ideas to large datasets and train on GPU, I highly recommend the GeomLoss library, which is optimized for this.
Instructions
Create a conda environment with all the requirements (edit environment.yml if you want to change the name of the environment):
conda env create -f environment.ymlActivate the environment
source activate pytorchOpen the notebook to reproduce the results:
jupyter notebook