yobx.sklearn.preprocessing.kernel_centerer

yobx.sklearn.preprocessing.kernel_centerer.sklearn_kernel_centerer(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: KernelCenterer, X: str, name: str = 'kernel_centerer') → str

Converts a sklearn.preprocessing.KernelCenterer into ONNX.

KernelCenterer centres a pre-computed kernel matrix K of shape (N, M) using statistics gathered from the training kernel matrix. The transformation implemented here mirrors sklearn.preprocessing.KernelCenterer.transform():

K_pred_cols = K.sum(axis=1, keepdims=True) / n_train   → (N, 1)

K_centered = K - K_fit_rows_              (broadcast: (N,M) - (M,))
               - K_pred_cols              (broadcast: (N,M) - (N,1))
               + K_fit_all_               (scalar)

where:

• K_fit_rows_ — column means of the training kernel matrix, shape (n_train,), stored as a constant initializer.

• K_fit_all_ — grand mean of the training kernel matrix, scalar.

• n_train — K_fit_rows_.shape[0], used to compute the per-sample row mean of the prediction kernel matrix.

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• outputs – desired output names

• estimator – a fitted KernelCenterer

• X – input tensor name (the kernel matrix K)

• name – prefix for added node names

Returns:

output tensor name