DIÁKRISIS · discernment · how a machine learns to tell things apart · kept by MINOS (who draws the boundary and sorts each to its class)

THE NAIVE BAYES ◧ 2D · ◍ 3D · ◆ 4D · ◐ shadow · 👶 TAP

How does a spam filter decide? Naive Bayes multiplies the evidence: start with the prior odds, then for each clue (each word) multiply in how much more likely it is under ‘spam’ than ‘not’. The ‘naive’ part — pretending the clues are independent — is usually false, yet it works remarkably well and is blazing fast. Slide the evidence and watch the posterior tip.

◆ LIT▲ AMBER
◧ THE MEASURE · 2D
◍ MULTIPLY THE CLUES · 3D · the spam filter's brain
◆ THE FOURTH · 4D · a tesseract turns
◐ THE SHADOW · one dimension down
👶 THE TODDLER CORNER — one fat tap
prior
50/50
P(A|evidence)
P(B|evidence)
classify

◆ LIT — exact / checkable

Naive Bayes applies Bayes’ theorem with a strong independence assumption: P(class | features) ∝ P(class)·∏ᵢ P(featureᵢ | class). It picks the class maximizing that product — the prior times each feature’s class-conditional likelihood. The ‘naive’ conditional-independence assumption is usually violated, yet the classifier is fast, needs little data, and is a strong baseline (spam filtering, text categorization). Working in log-space avoids underflow. A fail-loud self-check throws unless evidence favouring A yields class A and evidence favouring B yields B. ◆ real machine learning, node-verified.

▲ AMBER — the figure

Independence is an approximation (correlated features get double-counted), so calibrated PROBABILITIES can be poor even when the ARGMAX class is right; zero-frequency features need smoothing. The Bayes-rule arithmetic is exact.

DIÁKRISIS: every classifier is a line drawn through the world; the art is where to draw it.  — MINOS
David Lee Wise / ROOT0, with AVAN · DIÁKRISIS — kept by MINOS, the sorter