normalize_descriptor_embed_op_for_decaf

soda_analyzer.normalize_descriptor_embed_op_for_decaf(embed_op, embedding_dim=0, num_columns=0)

Normalize the persisted embedding operator to the DECAF model shape.

Parameters:

• embed_op (array-like) – Persisted descriptor embedding operator.

• embedding_dim (int, optional) – Target embedding dimension; zero permits inference from the operator. Default: 0.

• num_columns (int, optional) – Expected full descriptor width; zero permits inference. Default: 0.

Returns:

numpy.ndarray – Operator with dimensions consistent with the requested embedding.

Notes

HDF5 stores rows as embedding-operator columns, i.e. usually (num_columns, embedding_dim). Decaf reconstruction expects (embedding_dim, num_columns).