The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

importance instrument

DefinitionA per-feature score of how much a classifier reads it. For a tree it counts splits, which is the right sensitivity for a reader that is flat within leaves. Permutation importance and partial dependence answer related questions. Chapter 7 and chapter 14. Also permutation importance, partial dependence.
ExampleA tree that splits on x1 twice and on x3 once and never on x2 gives importances 2, 0, and 1.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id importance, kind instrument.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equationnone
Assumptions and scope
  • A per-feature score of how much a classifier reads it. For a gradient reader the averaged squared sensitivity is the read operator’s diagonal, and for a tree, whose finite difference is zero at every row that does not straddle a split, counting splits is the right sensitivity.
  • Permutation importance averages the consumer over rows that may never occur together, and partial dependence reads it at inputs it was never trained on, so the three answer different questions and the report names which.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/Attribution.lean, lean/DataMiningAsObservation/DecisionTree.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
x1x2x3x4x5x6
How much a classifier reads each feature, counting splits for a tree.

Equation

none

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Breiman, random forests, 2001, as chapter 7 section 7.1 and chapter 14 section 14.5 of Data Mining as Observation read it, with the selection-consumer regime in readscope/readscope/regimes.py:1-60.

Measurements

Where the book states it Numbers, as the book’s sources table records them Source
chapter 7 section 7.1 bagging variance, out-of-bag estimation, random forests, importances, AdaBoost Hastie, Tibshirani, Friedman, ESL 2e chapters 15 and 10.1; TSK 2e 4.10

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Attribution.lean, theorems attr_sum_affine, attr_unread, sq_sensitivity_eq_readOp_diag, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

lean/DataMiningAsObservation/DecisionTree.lean, theorems stump_flat_left, stump_flat_right, split_reads_one_axis, gini_le_half, gini_eq_zero_iff, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation chapters 0, 4, 6, 7, 14.

Related

attribution; decision tree; sensitivity; read operator; ensemble.

See also

Book equations stated beside the entry’s terms, not defining it: 14.5, 0.9, 6.1.

Ledger rows that cite the entry’s records without naming it: GO-EC-3.

Sources-table rows that share a record with the entry without naming it: chapter 2 section 2.3, chapter 6 section 6.1, chapter 14 section 14.5.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← identity readerimputation →