The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

margin concept

DefinitionHow far a row's score sits from the threshold. Chapter 6.
ExamplePoints at distance 0.5 and 0.8 from the boundary on either side give a margin of 0.5.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id margin, kind concept.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equation

Book equation 9.3.

Assumptions and scope
  • How far a row’s score sits from the threshold. A perturbation whose length times the weight length is smaller than the margin cannot move a linear classifier’s score across the threshold, by Cauchy–Schwarz, and a step along the weights of exactly that size reaches the boundary.
  • The margin is therefore a certificate on a decision, and chapter 9’s margin certificate is the same idea with a noise model in place of a perturbation bound and a vacuity threshold derived from it.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/Margin.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
marginboundary
The distance from the boundary to the nearest rows.

Equation

Book equation 9.3.

\[\hat\mu=\frac{\bar s_{\mathrm{true}}-\bar s_{\mathrm{distr}}}{\sigma_{\mathrm{distr}}},\qquad \mu_{\mathrm{crit}}=\mathbb E\Big[\max_{N-1}\mathcal N(0,1)\Big],\qquad \rho=\frac{\hat\mu}{\mu_{\mathrm{crit}}},\qquad \rho=1\ \text{vacuous}.\]

Conditions

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

Ledger

none

First stated

Chapter 6 section 6.1 of Data Mining as Observation, with the margin certificate of chapter 9 section 9.3 and the neighbour margins of chapter 11.

Measurements

none

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Margin.lean, theorems score_add, decision_stable, tight_along_weights, 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, 6, 7, 9, 11, 12, 14.

Related

classifier; decision boundary; vacuity threshold; rank certificate; escalation.

See also

Book equations stated beside the entry’s terms, not defining it: 11.2.

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

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.

← manifoldmatched bits →