Official Implementations of Deep Conditional Distribution Learning via Conditional Föllmer Flow

We present the implementations of our proposed conditional Föllmer flow in this repository, and examine its performance on 2 simulation studies and 3 real data analyses. We also compare the conditional Föllmer flow with existing NNKCDE, FlexCode, conditional GAN, conditional VAE, conditional VE-SDE and conditional Trigonometric flow. All the results in our manuscript can be reproduced with the slurm.sh in each sub-folder.

Dependencies

• pip install -r requirements
• You need to manual install nnkcde from https://github.com/lee-group-cmu/NNKCDE as its authors do not offer an official installation through PyPi or conda.

Hardware specification

We carry out the numerical experiments on 3 nodes of a NVIDIA 4xV100 cluster. In general, a machine with a 16GB NVIDIA GPU would satisfy the requirements for reproducing.

If you are using a personal desktop computer, you need to remove the slurm specificaitons in the slurm.sh and specify your python environments.

If you are using a slurm cluster, you might need to modify the slurm specifications such as the name of partition.