yobx.sklearn.model_selection.tuned_threshold_classifier_cv#

yobx.sklearn.model_selection.tuned_threshold_classifier_cv.sklearn_tuned_threshold_classifier_cv(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: TunedThresholdClassifierCV, X: str, name: str = 'tuned_threshold_classifier_cv') → Tuple[str, str][source]#

Converts a sklearn.model_selection.TunedThresholdClassifierCV into ONNX.

TunedThresholdClassifierCV wraps a binary classifier and adjusts the decision threshold for the positive class to optimise a scoring metric. At inference time it:

  1. Obtains the positive-class probability from the inner estimator via predict_proba.

  2. Compares that probability against best_threshold_.

  3. Returns the positive-class label when the probability is ≥ best_threshold_, and the negative-class label otherwise.

The ONNX graph replicates this logic:

X ──[estimator_ converter]──► (inner_label, probas (N, 2))
                                       │
                           Gather(axis=1, idx=pos_label_idx)──► y_score (N,)
                                       │
                           y_score >= best_threshold_ ──► bool (N,)
                                       │
                           Cast(INT64) ──► 0 or 1
                                       │
                   Gather(map_thresholded_score_to_label) ──► label_idx (N,)
                                       │
                   Gather(classes_) ──► label (N,)
Parameters:

  • g – the graph builder to add nodes to

  • sts – shapes and types defined by scikit-learn

  • outputs – desired output tensor names for the result

  • estimator – a fitted TunedThresholdClassifierCV

  • X – name of the input tensor

  • name – prefix used for names of nodes added by this converter

Returns:

tuple (label, probabilities)

Raises:

NotImplementedError – if the inner estimator does not expose predict_proba (i.e. response_method='decision_function' was used) — only probability-based thresholding is currently supported