API Reference

Stock Data Management

alphagen_qlib.stock_data.StockData

Constructor: __init__(self, instrument: Union[str, List[str]], start_time: str, end_time: str, ...)

  • instrument: The stock pool (e.g., 'csi300').
  • start_time: Start date string (YYYY-MM-DD).
  • end_time: End date string.
  • device: torch.device (CPU or CUDA).

Methods:

  • make_dataframe(data: torch.Tensor, columns): Converts the internal tensor representation back into a readable Pandas DataFrame with a MultiIndex (Date, Instrument).

Alpha Models

alphagen.models.linear_alpha_pool.LinearAlphaPool

Constructor: __init__(self, capacity: int, calculator: AlphaCalculator, ...)

  • capacity: Max number of alphas in the pool.
  • calculator: Instance of AlphaCalculator used for evaluating new alphas.

Methods:

  • try_new_expr(expr: Expression) -> float: Tests a new expression. If it improves the pool, it is added (and potentially triggers a prune). Returns the improvement metric.
  • optimize(): Recalculates weights for all alphas in the pool (usually via Ridge/Lasso regression).
  • test_ensemble(calculator): Returns (IC, RankIC) tuple for the entire pool on a specific dataset.

RL Environment

alphagen.rl.env.wrapper.AlphaEnv

Wraps the core environment with OpenAI Gym/Farama Gymnasium interfaces.

  • step(action): Takes an integer action ID.
  • reset(): Resets the expression builder.
  • action_masks(): Returns a boolean array indicating which tokens are valid for the next step (crucial for MaskablePPO).