---
license: bsd-3-clause
tags:
- atmospheric-retrieval
- exoplanets
- simulation-based-inference
pretty_name: SBI_INAF
size_categories:
- 10K<n<100K
---
# SBI_INAF

**SBI_INAF** is a dataset designed for **Simulation-Based Inference (SBI)** on atmospheric retrieval of exoplanets at a spectral resolution of approximately **R = 40000**, tailored for the GIANO-B echelle spectrograph. It contains transmission spectra (stellar and telluric free) together with their associated physical parameters, auxiliary metadata, normalization statistics, and predefined train/validation/test splits.

The dataset is built upon the Atmospheric Big Challenge (ABC) database, with major modifications to address the research goals of the research paper "Towards Scalable High-Resolution Atmospheric Retrieval of Exoplanets with Flow Matching" - M. Giordano Orsini et al., 2026.

The dataset is intended for supervised learning tasks involving parameter estimation from synthetic or simulated transmission spectra.

---

## Dataset Structure

```text
SBI_INAF_R1500/
├── AuxillaryTable.csv
├── FM_Parameter_Table.csv
├── h5s/
├── norm_params.pkl
├── QuartilesTable.csv
├── test_indices.npy
├── train_indices.npy
├── val_indices.npy
├── wavelengths.npy
├── widths.npy
```

---

## File Descriptions

| File | Description |
|------|-------------|
| `h5s/` | Collection of HDF5 files containing the spectra and associated sample information. |
| `FM_Parameter_Table.csv` | Physical parameters (forward-model parameters) associated with each simulated spectrum. These serve as prediction targets. |
| `AuxillaryTable.csv` | Additional metadata describing each sample (planetary system-related parameters). |
| `QuartilesTable.csv` | Summary statistics (quartiles) of selected parameters or features. Useful for exploratory data analysis. |
| `norm_params.pkl` | Stored normalization parameters used during preprocessing. |
| `wavelengths.npy` | Common wavelength grid for all spectra. |
| `widths.npy` | Spectral bin widths corresponding to the wavelength grid. |
| `train_indices.npy` | Indices defining the training split. |
| `val_indices.npy` | Indices defining the validation split. |
| `test_indices.npy` | Indices defining the held-out test split. |

---

## Citation

If this dataset contributes to published work, please cite:

```bibtex
@article{GiordanoOrsini_2026,
  author     = {Giordano Orsini, Massimiliano and Ferone, Alessio and Inno, Laura and Maratea, Antonio and Casolaro, Angelo and Giacobbe, Paolo and Pino, Lorenzo and Bonomo, Aldo S.}, 
  title      = {Towards Scalable High-Resolution Atmospheric Retrieval of Exoplanets with Flow Matching},
  year       = 2026,
  journal    = {Journal of Computational Physics},
  addendum   = {(Submitted)},
}
```

---

## License

Please refer to the accompanying repository or publication for licensing information.

---

## Contact

For questions regarding the dataset, please contact the dataset maintainers or the authors of the accompanying publication.