README.md

---
license: mit
pretty_name: AstroM3Dataset
size_categories:
- 10K<n<100K
tags:
- astronomy
- multimodal
- classification
arxiv:
- arXiv:2411.08842
---

# AstroM3Dataset

## Description

AstroM3Dataset is a time-series astronomy dataset containing photometry, spectra, and metadata features for variable stars. 
The dataset includes multiple subsets (`full`, `sub10`, `sub25`, `sub50`) and supports different random seeds (`42`, `66`, `0`, `12`, `123`). 
Each sample consists of:

- **Photometry**: Light curve data of shape `(N, 3)` (time, flux, flux\_error).
- **Spectra**: Spectra observations of shape `(M, 3)` (wavelength, flux, flux\_error).
- **Metadata**: Auxiliary features of shape `(38,)`.
- **Label**: The class name as a string.

## Corresponding paper and code

- Paper: [AstroM<sup>3</sup>: A self-supervised multimodal model for astronomy](https://arxiv.org/abs/2411.08842)
- Code Repository: [GitHub: AstroM<sup>3</sup>](https://github.com/MeriDK/AstroM3/)

---

## Subsets and Seeds
AstroM3Dataset is available in different subset sizes:  

- `full`: Entire dataset  
- `sub50`: 50% subset  
- `sub25`: 25% subset  
- `sub10`: 10% subset  

Each subset is sampled from the respective train, validation, and test splits of the full dataset.  
For reproducibility, each subset is provided with different random seeds:  

- `42`, `66`, `0`, `12`, `123`  


## Data Organization
The dataset is organized as follows:
```
AstroM3Dataset/
├── photometry.zip             # Contains all photometry light curves
├── utils/
│   ├── parallelzipfile.py     # Zip file reader to open photometry.zip
├── spectra/                   # Spectra files organized by class
│   ├── EA/
│   │   ├── file1.dat
│   │   ├── file2.dat
│   │   ├── ...
│   ├── EW/
│   ├── SR/
│   ├── ...
├── splits/                    # Train/val/test splits for each subset and seed
│   ├── full/
│   │   ├── 42/
│   │   │   ├── train.csv
│   │   │   ├── val.csv
│   │   │   ├── test.csv
│   │   │   ├── info.json      # Contains feature descriptions and preprocessing info
│   │   ├── 66/
│   │   ├── 0/
│   │   ├── 12/
│   │   ├── 123/
│   ├── sub10/
│   ├── sub25/
│   ├── sub50/
│── AstroM3Dataset.py          # Hugging Face dataset script
```

## Usage
To load the dataset using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

# Load the default full dataset with seed 42
dataset = load_dataset("MeriDK/AstroM3Dataset", trust_remote_code=True)
```

The default configuration is **full_42** (entire dataset with seed 42).
To load a specific subset and seed, use {subset}_{seed} as the name:

```python
from datasets import load_dataset

# Load the 25% subset sampled using seed 123
dataset = load_dataset("MeriDK/AstroM3Dataset", name="sub25_123", trust_remote_code=True)
```

---

## Citation
If you find this dataset usefull, please cite:
```bibtex
@article{rizhko2024astrom,
  title={AstroM $\^{} 3$: A self-supervised multimodal model for astronomy},
  author={Rizhko, Mariia and Bloom, Joshua S},
  journal={arXiv preprint arXiv:2411.08842},
  year={2024}
}
```