Skip to content
Quantoros

Why can a FeatureHasher not reliably recover the original feature names from transformed columns?

Classic MLLevel: IntermediateHashed feature inspection

Why can a FeatureHasher not reliably recover the original feature names from transformed columns?

Hints
  • Hashing is many-to-one.
  • No lookup table is retained.

Answer

It stores no vocabulary and multiple names can hash to the same index. This saves memory but loses exact inverse mapping and complicates coefficient interpretation.

Source and licence: scikit-learn 1.7.2 — feature extraction · BSD-3-Clause · answer: Quantoros editorial

Related questions