Janady13/missionmatch / internal/platform/learn/survival.go

Source identity

Field	Recorded value
Source type	GitHub source
Original record	Janady13/missionmatch / internal/platform/learn/survival.go
Date	Not specified in source
Repository	Janady13/missionmatch
Revision	720d47fae476b0d38477ae4401697d372b1ea2bc
Branch	feature/opportunity-to-outcome
Path or archive member	internal/platform/learn/survival.go
Original SHA-256	1ff30da0f0a6806e18436f5d3b5fde7b315be83340175295c8e8204500f5480e
Git blob SHA	97d97e65c9dbd7f59e58b8befece358600a8d77b
Reading-copy SHA-256	1ff30da0f0a6806e18436f5d3b5fde7b315be83340175295c8e8204500f5480e

Original source ↗

Every recorded occurrence

These 8 records contain identical original source bytes. Their paths and revisions remain part of the archive.

Original record	Branch or archive	Revision / identifier	Original location
Janady13/missionmatch / internal/platform/learn/survival.go	feature/opportunity-to-outcome	720d47fae476b0d38477ae4401697d372b1ea2bc	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	collision-deconfliction	4ee5a09972608ed89f8377c6665542d5f450f0a6	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	master	1e9249840a4302ec82e123b9c9cb6f75badea3e4	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	time-to-fill-survival	b66e1e505863d4f17741a4b9ede73509e45bd6c6	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	presentation-readiness	eb574acaa29d71ce4d0f852572e87c275c79fdb7	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	vendor-intelligence	fe9796a19c261951d3f2565206c0c1b5ac294465	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	feature/aioe-upgrade	2178e7298d0e5d9af83afa12402fd587f0ba6ed8	Open original
Janady13/missionmatch / internal/platform/learn/survival.go	capture-readiness	bae6af7cb2997e819174a9e9cfff5e16067a1975	Open original

Recovered source transcript

Download complete source text ↓100 lines · 3.7 KB

package learn

// Discrete time survival for the time to fill forecast: how long a staffing requirement stays open before it is
// filled. It is a Kaplan Meier style discrete hazard, fully interpretable and pure: a per period (month) hazard
// is estimated from right censored durations, the cumulative survival is the running product of one minus the
// hazard, and the probability of filling within a horizon is one minus the survival at that horizon. A still
// open requirement is right censored: it is known to have survived to its observed period without filling, but
// not when it will fill, so it counts in the at risk denominator through its observed periods and never in the
// event numerator. No database, no Python; float lives only in this offline estimator and the caller quantizes
// every stored and surfaced number to decimal.

// Duration is one requirement's time to fill observation: the whole periods (months) observed, and whether the
// fill event occurred at the last observed period (Filled true) or the observation is right censored, still open
// at the last period (Filled false). Periods is at least 1.
type Duration struct {
	Periods int
	Filled  bool
}

// EmpiricalHazard estimates the discrete time hazard per period from right censored durations. For each period m
// (1 based, index m-1 in the returned vector), hazard[m] = events at m / at risk at m, where at risk at m counts
// every observation whose Periods is at least m (it reached period m still open at the start of it), and events
// at m counts the observations that Filled exactly at period m. A still open (censored) observation contributes
// to every at risk denominator through its observed periods but to no event numerator. The returned vector has
// length maxPeriod; a period with no one at risk has hazard 0 (no evidence of filling then). maxPeriod at most 1
// returns a single period.
func EmpiricalHazard(durations []Duration, maxPeriod int) []float64 {
	if maxPeriod < 1 {
		maxPeriod = 1
	}
	events := make([]float64, maxPeriod)
	atRisk := make([]float64, maxPeriod)
	for _, d := range durations {
		p := d.Periods
		if p < 1 {
			p = 1
		}
		if p > maxPeriod {
			p = maxPeriod
		}
		// At risk in every period up to and including its observed period.
		for m := 0; m < p; m++ {
			atRisk[m]++
		}
		// The event (a fill) lands in the observed period only when it filled, not when it is censored.
		if d.Filled {
			events[p-1]++
		}
	}
	hazard := make([]float64, maxPeriod)
	for m := 0; m < maxPeriod; m++ {
		if atRisk[m] > 0 {
			hazard[m] = events[m] / atRisk[m]
		}
	}
	return hazard
}

// Survival returns the cumulative survival curve from a per period hazard vector: S[t] = product over i in 0..t
// of (1 - hazard[i]). S has the same length as hazards; S[t] is the probability of still being open (not yet
// filled) at the end of period t+1. A hazard clamped outside [0,1] is treated as its nearest bound so the
// survival stays a valid probability.
func Survival(hazards []float64) []float64 {
	s := make([]float64, len(hazards))
	running := 1.0
	for t, h := range hazards {
		if h < 0 {
			h = 0
		}
		if h > 1 {
			h = 1
		}
		running *= (1 - h)
		s[t] = running
	}
	return s
}

// FillWithin returns the probability the requirement is filled within `horizon` periods: 1 - S(horizon). A
// horizon at or below zero is 0 (nothing fills in no time); a horizon beyond the hazard vector uses the full
// vector (the survival cannot fall further with no more hazard evidence). The result is a probability in [0, 1].
func FillWithin(hazards []float64, horizon int) float64 {
	if horizon <= 0 || len(hazards) == 0 {
		return 0
	}
	s := Survival(hazards)
	idx := horizon - 1
	if idx >= len(s) {
		idx = len(s) - 1
	}
	p := 1 - s[idx]
	if p < 0 {
		return 0
	}
	if p > 1 {
		return 1
	}
	return p
}

Mathematical records from this source

survival · (1 based, index m-1 in the returned vector), hazard[m]

This relation defines (1 based, index m-1 in the returned vector), hazard[m] in the source section “survival · (1 based, index m-1 in the returned vector), hazard[m]”. It…

Survival probability from conditional hazards

Each hazard is the conditional probability of an event in that period given survival up to that period. Multiplying successive complements yields the probability of remaining…