yobx.sklearn.preprocessing.normalizer

yobx.sklearn.preprocessing.normalizer.sklearn_normalizer(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: Normalizer, X: str, name: str = 'normalizer') → str

Converts a sklearn.preprocessing.Normalizer into ONNX.

Each sample (row) is scaled independently to unit norm. The norm parameter selects which norm is used:

norm = ReduceL2/L1/Max(|X|, axis=1, keepdims=True)   → (N, 1)
safe_norm = Where(norm == 0, 1, norm)                  → (N, 1)
output = X / safe_norm                                 → (N, F)

When a row has zero norm it is left unchanged (divided by 1), which matches sklearn.preprocessing.normalize().

Supported values for norm: 'l2' (default), 'l1', 'max'.

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• estimator – a fitted or unfitted Normalizer

• outputs – desired output names

• X – input tensor name

• name – prefix for added node names

Returns:

output tensor name