yobx.sklearn.decomposition.sparse_pca

yobx.sklearn.decomposition.sparse_pca.sklearn_sparse_pca(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: SparsePCA | MiniBatchSparsePCA, X: str, name: str = 'sparse_pca') → str

Converts a sklearn.decomposition.SparsePCA or sklearn.decomposition.MiniBatchSparsePCA into ONNX.

The transformation replicates sklearn.decomposition.SparsePCA.transform():

1. Centre the data by subtracting the per-feature mean stored in mean_.

2. Apply the pre-computed projection matrix W via a single matrix multiplication, where W is derived from the fitted dictionary components_ and the regularisation parameter ridge_alpha:

with = components_ and = ridge_alpha.

The full graph is:

X  ──Sub(mean_)──►  centered  ──MatMul(W)──►  output

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• estimator – a fitted SparsePCA or MiniBatchSparsePCA

• outputs – desired output names (sparse codes)

• X – input tensor name

• name – prefix name for the added nodes

Returns:

output tensor name