yobx.sklearn.multioutput.multioutput#

yobx.sklearn.multioutput.multioutput.sklearn_multi_output_classifier(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: MultiOutputClassifier, X: str, name: str = 'multi_output_classifier') → str | Tuple[str, str][source]#

Converts a sklearn.multioutput.MultiOutputClassifier into ONNX.

The converter iterates over the fitted sub-estimators, calls the registered converter for each one to obtain per-target labels, and concatenates the reshaped (N, 1) labels into a final label tensor of shape (N, n_targets).

When probabilities are requested (i.e. len(outputs) > 1) and every target has the same number of classes, the per-target probability matrices (N, n_classes) are unsqueezed to (N, 1, n_classes) and concatenated into a (N, n_targets, n_classes) tensor.

Graph structure (labels only):

X ──[sub-est 0 converter]──► label_0 (N,) ──Cast(INT64)──Reshape(N,1)──┐
X ──[sub-est 1 converter]──► label_1 (N,) ──Cast(INT64)──Reshape(N,1)──┤
...                                                                    │
                 +-----------------------------------------------------+
                 │
Concat(axis=1) ─────────► labels (N, n_targets)

Graph structure (with probabilities, all targets same n_classes):

X ──[sub-est 0 converter]──► proba_0 (N, n_cls) ──Unsqueeze──► (N,1,n_cls)──┐
X ──[sub-est 1 converter]──► proba_1 (N, n_cls) ──Unsqueeze──► (N,1,n_cls)──┤
...                                                                         │
                      +-----------------------------------------------------+
                      │
    Concat(axis=1) ───────────► probabilities (N, n_targets, n_cls)
Parameters:

  • g – the graph builder to add nodes to

  • sts – shapes and types defined by scikit-learn

  • outputs – desired output tensor names (label, or label + probabilities)

  • estimator – a fitted MultiOutputClassifier

  • X – name of the input tensor

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

Returns:

label tensor name, or tuple (label, probabilities)

Raises:

NotImplementedError – when probabilities are requested but targets have a different number of classes or sub-estimators do not expose predict_proba()

yobx.sklearn.multioutput.multioutput.sklearn_multi_output_regressor(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: MultiOutputRegressor, X: str, name: str = 'multi_output_regressor') → str[source]#

Converts a sklearn.multioutput.MultiOutputRegressor into ONNX.

The converter iterates over the fitted sub-estimators, calls the registered converter for each one, and reshapes every 1-D prediction (N,) to (N, 1) before concatenating them into the final output of shape (N, n_targets).

Graph structure:

X ──[sub-est 0 converter]──► pred_0 (N,) ──Reshape(N,1)──┐
X ──[sub-est 1 converter]──► pred_1 (N,) ──Reshape(N,1)──┤
...                                                        │
                                      Concat(axis=1) ─────► predictions (N, n_targets)
Parameters:

  • g – the graph builder to add nodes to

  • sts – shapes and types defined by scikit-learn

  • outputs – desired output tensor names

  • estimator – a fitted MultiOutputRegressor

  • X – name of the input tensor

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

Returns:

name of the output tensor of shape (N, n_targets)