Skip to content

Latest commit

 

History

History
103 lines (73 loc) · 4.2 KB

README.md

File metadata and controls

103 lines (73 loc) · 4.2 KB

We introduce Ex🪐plain, a model-agnostic framework to identify and describe exoplanetary characteristics that challenge atmospheric parameterization.

Papers

Code associated with the papers

Data

We do not include the dataset used in the paper as it is publicly available and downloadable from the ARIEL Big Data Challenge official site.

Get Started

Our code was tested on Python 3.10.4. To make it work, you will need a working environment with the libraries listed in requirements.txt.

Running the Experiments

To reproduce the paper's experiments, you can directly run the exoplain.ipynb notebook, which leverages the predictions already computed for both the ensemble and the weak learners.

What you can do

Individual model analysis

You can investigate the sources of errors for a model (e.g., the ensemble), and highlight the subgroups with the highest negative (or positive) divergence, as shown here:

And plot the Shapley values of these subgroups to understand the impact of each single item on the performance of the model:

Model comparison analysis

You can also investigate the gain in performance at the subgroup level when changing one model (e.g., a weak learner) with another one (e.g., an ensemble). For example, the following table shows the subgroups that most benefit or fail to benefit when changing these two models:

And plot the distribution of gain in performance for all the retrieved subgroups (cross-hatched green denotes performance improvement, while red indicates performance decrease):

Global divergence analysis

Finally, you can exploit the global Shapley values to investigate the impact of the items on the performance of the model. For example, the following image shows the top-15 items that have the highest impact on the performance for the ensemble (cross-hatched green denotes performance improvement, while red indicates performance decrease):

Citation

If you use this code in your research, please cite our papers:

@ARTICLE{10689522,
  author={Koudounas, Alkis and Giobergia, Flavio and Baralis, Elena},
  journal={IEEE Access}, 
  title={Ex(o)plain: Subgroup-level analysis of exoplanet atmospheric parameters}, 
  year={2024},
  volume={},
  number={},
  pages={1-1},
  keywords={Atmospheric modeling;Exoplanet;Atmospheric measurements;Computational modeling;Atmospheric waves;Metadata;Deep learning;Explainable AI;Detection algorithms;Deep Learning;Divergence;Exoplanet Atmospheric Parameters;Explainable AI;Subgroup Detection},
  doi={10.1109/ACCESS.2024.3466919}}
@inproceedings{koudounas2023bad,
  title={Bad Exoplanet! Explaining Degraded Performance when Reconstructing Exoplanets Atmospheric Parameters},
  author={Alkis Koudounas and Flavio Giobergia and Elena Baralis},
  booktitle={NeurIPS 2023 AI for Science Workshop},
  year={2023},
  url={https://openreview.net/forum?id=9Z4XZOhwiz}
}

License

This code is released under the Apache 2.0 license. See the LICENSE file for more details.

Contact

For any questions, please contact Alkis Koudounas.