---
license: cc-by-4.0
tags:
  - hackathon
  - alphahack
  - factor-model
---

# AlphaHack Trained Models

Three trained scikit-learn artifacts for the AlphaHack quantitative
hackathon-strategy engine. Each lives in its own subdirectory with a
dedicated model card.

| Subdirectory | Artifact | Purpose | Size |
|---|---|---|---|
| `model1-regime-classifier/` | `regime_classifier.pkl` | Event → winner-archetype classifier (Model 1) | 2.2 MB |
| `model2-winner-predictor/` | `winner_predictor_final.pkl` | Project → ex-ante prize-probability classifier (Model 2) | 493 KB |
| `text-embedder/` | `text_embedder.pkl` | TF-IDF + TruncatedSVD text embedder (auxiliary) | 4.2 MB |

Companion artifacts:
- **Dataset**: [`xenosaac/alphahack-devpost`](https://huggingface.co/datasets/xenosaac/alphahack-devpost)
- **Source code**: [`xenosaac/Alpha-Hack`](https://github.com/xenosaac/Alpha-Hack)

## Honest framing

These models were validated retrospectively (Model 2 sponsor-prize AUC
0.908, 95% CI [0.859, 0.947]) and tested in **one prospective trial in
April 2026 that did not produce a prize**. Treat as a research artifact,
not a guaranteed winning strategy. Each model card documents the known
failure modes specific to that artifact.

## Quickstart

```bash
pip install hackalpha
python -c "
import joblib
m2 = joblib.load('model2-winner-predictor/winner_predictor_final.pkl')
print(m2['features'])  # the 23 features it expects
"
```

The `text-embedder/text_embedder.pkl` requires the `hackalpha` package
to be importable at load time (it pickles a `hackalpha.research.text_embeddings.TextEmbeddingFeatures`
instance). The other two models are pure scikit-learn and load without
any project-specific deps.

## License

CC BY 4.0 — see `LICENSE`.