---
license: mit
tags:
  - quantitative-finance
  - alpha-research
  - backtesting
  - kaggle
---

# PuLam/robust-AAE

Lightweight standalone bundle for **AAE alpha robustness testing**.

This bundle is intentionally scoped to the robustness path only. It does **not**
overwrite or modify the stable backtest flow.

## Included

- robust standalone runner
- robustness matrix runner
- CSV -> JSONL alpha-pack converter
- top-10 alpha input pack exported from `test7`
- small sample outputs for one alpha and a small matrix smoke run
- Kaggle offline instructions

## Key scripts

- `scripts/jsonl_alpha_robustness.py`
- `scripts/run_alpha_robustness_matrix.py`
- `scripts/prepare_alpha_robustness_input.py`
- `scripts/publish_alpha_robustness_to_hf.py`

## Kaggle offline usage

The recommended runtime source remains the Kaggle dataset `gplebih/aae-new`, because it
already includes the backtest runtime, wheelhouse, and `daily_pv.h5`.

See:

- `docs/README_alpha_robustness.md`
- `docs/README_alpha_robustness_matrix.md`

## Top-10 alpha pack

- `inputs/test7_top10_best_return_alpha_pack.jsonl`
- `inputs/top10_seed_best_return_with_formula.csv`

## Sample outputs

- `examples/decay_relret_mom_v4_baseline/`
- `examples/decay_relret_mom_v4_matrix_smoke/`