ExoMiner v2.1 and Podman release 🚨

[migmartinho]

Hi! 👋

It has been a while, but we have updated the ExoMiner codebase to v2.1 - including the Podman image with an improved ExoMiner Pipeline! We advise people to read this.

Instructions on how to set up ExoMiner Pipeline with Podman: here.

Major Changes

• The pipeline now accepts two types of classification tasks: 3-class photometric vetting (PC vs AFP vs NTP) and 2-class planet vs not-planet.
• Both tasks can be run using a single model or two ensemble variants.
• Change the architecture of the pipeline in terms of how parallelization works for both the data extraction/preprocessing step and the model inference step (see here).
• For the ensemble variants, the standard deviation of the ensemble is reported in the prediction table as an optional output that provides the user with insights into the model's uncertainty about a given TCE.
• 📈An additional output can be provided for each TCE by using the flag plot_inputs_to_model. This will generate a summary report that shows the preprocessed inputs like flux time series and difference images provided to the model for the TCE. See here for more information.
• The same shell script can be used for running the pipeline using the Podman image or using Python. See here for more details.

Minor Changes

• For reproducibility and tracking/versioning, the ExoMiner models (plus normalization statistics) were added both to Zenodo and Hugging Face. Future model versions will be added here.
• In here there is a detailed "model card" for the models currently shipped with the ExoMiner Pipeline.
• TCE and TIC summaries CSVs are output by the pipeline so the user can see whether they were successfully processed and if they were not, then they can see more easily where it failed (still check logs to get the actual errors as well as look at the plots if they are generated).

Happy planet hunting! 🔭🛰️