Fit

object Fit(source)

Properties

LAPLACE

const val LAPLACE: Int = 1

NONE_GRID

val NONE_GRID: List<Long>

Weights of "no listed candidate is the cause" tried by noneWeight, micro-units.

UNIT_NONE

const val UNIT_NONE: Long

The weight of "none" when nothing is fit: the same as having no evidence.

Functions

count

fun count(cands: List<Cand>): Count

evidence

fun evidence(train: List<Observed>, base: Table = Table.handSet): Table

The probe ratio (a differing probe against none) and one rule ratio shared by every severity (a rule match against none), with the dimension weights of base; there are too few matches per severity. Evidence that never occurs in train keeps the weight of base.

full

fun full(train: List<Observed>, base: Table = Table.handSet): Table

Priors plus evidence.

inputs

fun inputs(o: Observed, t: Table, normalize: Boolean, none: Long = UNIT_NONE): List<Long>

What the calibration map reads for each candidate of o: its summed log-odds, or with normalize its share of the posterior against its rivals and "no listed candidate is the cause" (weight 1), so a scenario with many changes gives each one less.

isotonic

fun isotonic(train: List<Observed>, table: Table, normalize: Boolean): Isotonic

noneWeight

fun noneWeight(train: List<Observed>, t: Table): Long

The grid value with the lowest training Brier score of the softmax share, over the rankings that get a printed probability (Observed.covered).

priors

fun priors(train: List<Observed>, base: Table = Table.handSet): Table

Per-dimension prior only; probe and rule weights keep the values of base.