yobx.sklearn.decomposition.fast_ica

yobx.sklearn.decomposition.fast_ica.sklearn_fast_ica(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: FastICA, X: str, name: str = 'fast_ica') → str

Converts a sklearn.decomposition.FastICA into ONNX.

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

• When whitening was applied during fitting (estimator.whiten is not False), the per-feature mean stored in mean_ is subtracted first.

• The data (centred or raw) is then projected onto the independent components via a matrix multiplication with components_.T.

X  ──Sub(mean_)──►  centered  ──MatMul(components_.T)──►  output
     (only when whiten != False)

When whiten is False the Sub node is omitted and the unmixing matrix is applied directly to X.

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• estimator – a fitted FastICA

• outputs – desired output names (independent components)

• X – input tensor name

• name – prefix name for the added nodes

Returns:

output tensor name