/tropical_cyclone_uq

Code to reproduce experiments for the paper

Primary LanguageJupyter NotebookMIT LicenseMIT

tropical_cyclone_uq

Code for paper: "Uncertainty Aware Tropical Cyclone Wind Speed Estimation from Satellite Data" at ICLR Workshop Machine Learning for Remote Sensing 2024.

Installation

pip install lightning-uq-box
pip install torchgeo
pip install wandb

Run Experiments

The run_cli_script.py executes the experiment and expects a config file.

For example to run a BNN model.

python run_cli_script.py model_config=configs/cyclone_reg/bnn_elob.yaml data_config=configs/cyclone_reg/dataset.yaml trainer_config=configs/cyclone_reg/trainer.yaml experiment.seed=63 trainer.devices=[0]`

Analysis

The plots in the paper are all generated with the results_tropical.ipynb notebook that expects a directory where the experiments are stored.