/deconditional-downscaling

This repo contains materials for the "Deconditional Dowscaling with Gaussian Processes" NeurIPS 2021 paper

Primary LanguagePython

Deconditional Downscaling with Gaussian Processes

figure

Getting started

  • Run Swiss Roll experiment with matched dataset

Pick one of the configuration files listed under

experiments/swiss_roll/config/
├── exact_cmp.yaml
├── bagged_gp.yaml
├── variational_cmp.yaml
├── vbagg.yaml
└── gp_regression.yaml

and run from root directory

$ python experiments/swiss_roll/run_experiment.py --cfg=path/to/config/file --o=path/output/directory
  • Run Swiss Roll experiment with unmatched dataset

Simply add --unmatched option to the above as

$ python experiments/swiss_roll/run_experiment.py --cfg=path/to/config/file --o=path/output/directory --unmatched
  • Run Downscaling experiment

Pick one of the configuration files listed under

experiments/downscaling/config/
├── variational_cmp_indiv_noise.yaml
├── vbagg.yaml
└── krigging.yaml

and run from root directory

$ python experiments/downscaling/run_experiment.py --cfg=path/to/config/file --o=path/output/directory

TODO

  • Update requirements
  • Sanity check experiments

Installation

Code implemented in Python 3.8.0

Setting up environment

Create and activate environment

$ pyenv virtualenv 3.8.0 venv
$ pyenv activate venv
$ (venv)

Install dependencies

$ (venv) pip install -r requirements.txt

Reference

@inproceedings{ChaBouSej2021,
  title = {{Deconditional Downscaling with Gaussian Processes}},
  author = {Chau, Siu Lun and Bouabid, Shahine and Sejdinovic, Dino},
  year = {2021},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)}
}