README.md

---
license: apache-2.0
tags:
  - tabular
  - synthetic-data
viewer: true
configs:
  - config_name: flat
    data_files:
    - split: train
      path: data/flat/train/*.csv
  - config_name: sequential
    data_files:
    - split: train
      path: data/sequential/train/*.csv
---

# MOSTLY AI Prize Dataset

This repository contains the dataset used in the [MOSTLY AI Prize](https://www.mostlyaiprize.com/) competition.

## About the Competition

Generate the BEST tabular synthetic data and win 100,000 USD in cash.
Competition runs for 50 days: May 14 - July 3, 2025.

This competition features two independent synthetic data challenges that you can join separately:

1. The FLAT DATA Challenge
2. The SEQUENTIAL DATA Challenge

For each challenge, generate a dataset with the same size and structure as the original, capturing its statistical patterns — but without being significantly closer to the (released) original samples than to the (unreleased) holdout samples.

Train a generative model that generalizes well, using any open-source tools ([Synthetic Data SDK](https://github.com/mostly-ai/mostlyai), [synthcity](https://github.com/vanderschaarlab/synthcity), [reprosyn](https://github.com/alan-turing-institute/reprosyn), etc.) or your own solution. Submissions must be fully open-source, reproducible, and runnable within 6 hours on a standard machine.

## Timeline

- Submissions open: May 14, 2025, 15:30 UTC
- Submission credits: 3 per calendar week (+bonus)
- Submissions close: July 3, 2025, 23:59 UTC
- Evaluation of Leaders: July 3 - July 9
- Winners announced: on July 9 🏆

## Dataset Description

This dataset consists of two CSV files used in the MOSTLY AI Prize competition:

### Flat Data
- File: `data/flat/train/flat-training.csv` (26MB, MD5 `d5642dd9b13da0dc1fbac6f92f8e4b20`)
- 100,000 records
- 80 data columns: 60 numeric, 20 categorical

### Sequential Data
- File: `data/sequential/train/sequential-training.csv` (6.6MB, MD5 `dd024fe8130cb36ad9374e23ccbffc4a`)
- 20,000 groups
- Each group contains 5-10 records
- 11 data columns: 7 numeric, 3 categorical + 1 group ID

### Data Format

You can load them directly using `pandas`:

```python
import pandas as pd

flat_df = pd.read_csv('data/flat/train/flat-training.csv')
sequential_df = pd.read_csv('data/sequential/train/sequential-training.csv')
```

Or using Hugging Face's `datasets`:

```python
from datasets import load_dataset

flat_dataset = load_dataset("mostlyai/mostlyaiprize", "flat", split="train")
sequential_dataset = load_dataset("mostlyai/mostlyaiprize", "sequential", split="train")
```

## Dataset Schema

The schema of each dataset can be retrieved as follows:

```python
# pandas
print(flat_df.dtypes)
print(sequential_df.dtypes)

# HF datasets
print(flat_dataset.features)
print(sequential_dataset.features)
```

### Column Description

Note: Detailed column descriptions are intentionally not provided as part of the competition challenge. The task is to generate synthetic data that preserves the statistical properties of the original data without needing to understand the semantic meaning of each column.

### Notes on Holdout Data

The competition evaluates submissions against a hidden holdout set that:
- Has the same size as the training data
- Does not overlap with the training data
- Comes from the same source
- Has the same structure and statistical properties

Your synthetic data generation approach should generalize well to this unseen data.

## Evaluation

- CSV submissions are parsed using `pandas.read_csv()` and checked for expected structure & size
- Evaluated using the [Synthetic Data Quality Assurance](https://github.com/mostly-ai/mostlyai-qa) toolkit
- Compared against the released training set and a hidden holdout set (same size, non-overlapping, from the same source)

## Citation

If you use this dataset in your research, please cite:

```
@dataset{mostlyaiprize,
  author = {MOSTLY AI},
  title = {MOSTLY AI Prize Dataset},
  year = {2025},
  url = {https://www.mostlyaiprize.com/},
}
```

## License

This dataset is provided under the Apache License 2.0. See the LICENSE file for full licensing information.