yobx.sklearn.kernel_approximation.rbf_sampler

yobx.sklearn.kernel_approximation.rbf_sampler.sklearn_rbf_sampler(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: RBFSampler, X: str, name: str = 'rbf_sampler') → str

Converts a sklearn.kernel_approximation.RBFSampler into ONNX.

The transform replicates sklearn.kernel_approximation.RBFSampler.transform() using random Fourier features (Random Kitchen Sinks) to approximate the RBF (Gaussian) kernel:

X_new = sqrt(2 / n_components) * cos(X @ random_weights_ + random_offset_)

where random_weights_ has shape (n_features, n_components) and random_offset_ has shape (n_components,).

Graph structure:

X ──MatMul(random_weights_)──Add(random_offset_)──Cos──Mul(scale)──► X_new

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• estimator – a fitted RBFSampler

• outputs – desired output names (transformed inputs)

• X – input tensor name (shape (N, F))

• name – prefix name for the added nodes

Returns:

output tensor name (shape (N, n_components))