yobx.sklearn.feature_extraction.tfidf_transformer

yobx.sklearn.feature_extraction.tfidf_transformer.sklearn_tfidf_transformer(g: GraphBuilderExtendedProtocol, sts: Dict, outputs: List[str], estimator: TfidfTransformer, X: str, name: str = 'tfidf_transformer') → str

Converts a sklearn.feature_extraction.text.TfidfTransformer into ONNX.

The transformer applies the following steps in order:

1. Term-frequency scaling (sublinear_tf): if True, replace each non-zero count with 1 + log(count); zero counts stay zero.

2. IDF weighting (use_idf): if True, multiply each term-frequency value element-wise by the fitted idf_ vector.

3. Row normalisation (norm): scale each row to unit 'l2' or 'l1' norm; None skips this step.

Graph layout (all three options active):

X  ──Greater(0)──────────────────────────────────┐
   ──Log ──────── Add(1) ──── Where(>0, ·, 0) ───┤
                                                  Mul(idf_) ──ReduceL2──Div── output

Parameters:

• g – the graph builder to add nodes to

• sts – shapes defined by scikit-learn

• estimator – a fitted TfidfTransformer

• outputs – desired output names

• X – input tensor name (shape (N, F), dtype float32 or float64)

• name – prefix for added node names

Returns:

output tensor name