cleverly.learners.SuperLearnerDiagnostics#

class cleverly.learners.SuperLearnerDiagnostics(names: tuple[str, ...], cv_risk: ndarray[tuple[Any, ...], dtype[float64]], weights: ndarray[tuple[Any, ...], dtype[float64]], best: str, ensemble_cv_risk: float, loss: str, failed: tuple[str, ...] = ())[source]#

What the ensemble learned, for reporting and validation.

Parameters:

namestuple of str: Candidate names, in the order the weights refer to.
cv_riskndarray: Cross-validated risk per candidate (mean squared error for a regression task, negative log-likelihood for a classification task).
weightsndarray: Convex ensemble weights.
beststr: Name of the lowest-risk candidate – what the discrete Super Learner would have selected.
ensemble_cv_riskfloat: Cross-validated risk of the weighted ensemble itself. Comparing it with min(cv_risk) shows whether stacking bought anything.
lossstr: Name of the loss the risks above were scored under.
failedtuple of str: Candidates dropped because fitting raised.

Methods#

SuperLearnerDiagnostics.to_dict() → dict[str, Any][source]#

Return a JSON-compatible representation.

Returns:

dict of str to Any: Candidate names, cross-validated risks, and weights, as JSON-compatible lists.