This is the official implementation of PICProp: Physics-Informed Confidence Propagation for Uncertainty Quantification (NeurIPS 2023).
Arxiv: https://arxiv.org/abs/2310.06923
Please feel free to shoot an email to shenqianli@u.nus.edu for any questions.
Install required python packages:
pip install -r requirements.txt
Add the your/path/to/PICProp
to $PYTHONPATH
:
export PYTHONPATH=$PYTHONPATH:your/path/to/PICProp
Enter desired experiment directory, for example:
cd your/path/to/PICProp/experiment/pedagogical
Execute scripts to start training. For PICProp in a brute-force manner:
python run_picprop.py
For EffiPICProp,
python run_effipicprop.py --lamb $lamb
The script will automatically sweep required query points and start a PICProp run if the result is absent, and summarize the result in a basic plot.