ax.health_check

Ax Experiment Health Checks

Search Space

ax.health_check.search_space.search_space_update_recommendation(search_space: SearchSpace, parametrizations: list[dict[str, None | str | bool | float | int]], boundary_prop: float = 0.5, tol: float = 1e-06) tuple[dict[str, tuple[float, float]], str][source]

Recommendation to update the search space boundaries depending on the generated candidates. If most of them fall on some of the boundaries, we suggest expanding the search space along those boundaries.

Note that this function does not handle parameter constraints. TODO: Add support for search space update in the presence of paramater constraints.

Parameters:
  • search_space – Search space.

  • parametrizations – A list of suggested parametrizations (parameter values).

  • boundary_prop – The minimal proportion of suggested parametrizations that land at the boundary of the search space for us to recommend expanding the search space.

  • tol – Tolerance for the difference between parameters and the boundary lower and upper bound.

Returns: A tuple consisting of
  • a dictionary mapping parameter names into the proportion of suggested

    parameterizations that landed on the lower and upper bound of that parameter respectively and

  • a human readable string containing the recommendation on the

    search space expansion to expand the boundaries where more than a specified proportion of parametrizations land.