Open In Colab

DeepFunding Mini is a mini-contest to create a distilled human judgment mechanism predicting the amount of past funding received by open source projects.

The mini dataset represents pairs of GitHub repositories taken from the Ethereum dependency graph and their funding amounts in historical funding rounds data.

• Training Data: 3,410 comparisons between 245 repos, 2,387 provided for training
• Test Data: 1,023 predictions on the relative funding received between any 2 projects, which get tested against actual data

Dataset: evalscience/deepfunding-mini

Predict the relative funding received between pairs of projects. Each comparison should output a relative value between repositories that sums to 1.

Winners receive up to 20,000ドル based on:

• Novelty and approach in predicting 1,023 comparisons
• Infrastructure and datasets created

Evaluation Metric: Mean Squared Error (MSE) - lower is better, but avoid overfitting.

Important: Submission weights must maintain transitivity:

• For any triple a, b, c: c/a = c/b * b/a
• Example: If A gets twice as much as B, and B gets three times as much as C, then A must be six times better than C

Join us at Eval.Science to become part of the next generation of decentralized data scientists tackling important problems like DeepFunding.

• 🎥 YouTube Channel
• 📚 Lecture Notes (WIP)
• 🤗 Hugging Face Space
• 📊 GitHub Dependency Graph
• 💻 Eval.Science Website

1. Click the "Open in Colab" button above
2. Make a copy of the notebook
3. Run all cells
4. Start experimenting!

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

@misc{deepfunding2024,
 title={DeepFunding Mini Dataset: Predictive Funding Challenge for Open Source Dependencies},
 author={DeepFunding.org, Eval.Science},
 year={2024},
 publisher={Hugging Face},
 howpublished={\url{https://huggingface.co/datasets/evalscience/deepfunding-mini}}
}

This project is licensed under the MIT License - see the LICENSE file for details.