The adaptation proposed for Foreseen is to deduce again with the forecasts, from what was done or from scratch (alert box of the supervision).
Prospective distance
The other strategies compare the plan with the field as it is now. This one compares it with the field as the forecasts foresee it: distance(EXPECTED, ACTIVE + PROJECTED(t)). A problem (actuality, potentiality, risk) that a forecast projects and that no objective of the supervised solution treats or prevents raises the Foreseen category before it reaches the field.
implemented byProspectiveDistanceStrategy · gind-npm-strategy-supervision-prospective-distance
01 · Projections
What the forecasts foresee
From the forecast projection — at each computation the supervision engine adds to the context (projections) the last projection of each forecast of the knowledge space, read from the forecast projection service. A projection gives, for each projected node, its distribution over the dates of the forecast: at each date, the share of the forecast curves on which the node is ACTIVE (active) and, for a risk, the share on which its probability reaches 1 (certain).
No forecast, no value — when the knowledge space has no projection the strategy returns nothing and Foreseen keeps no value from it. Dates are read on the clock of the use case, like every supervision computation: the past points of a projection only help to know when a problem appears, the dates judged are those after now.
02 · Problems foreseen
Which projected nodes count
A problem the plan ignores — a projected node counts when its concept (its own type, then its ancestors in the collaborative model) is Actuality, Potentiality or Risk, when the solution does not treat or prevent it, and when it is not already ACTIVE on the field: a problem already there is the business of Uncovered, a projected objective is not a problem.
Its share and the date it appears — the share of a node is its active share, or for a risk the larger of active and certain. It appears at the first date its share reaches the “Counted from” share; a node whose share never reaches it, or whose weight is 0, is ignored.
03 · Score
Weight × share × urgency
value(t) = Σ weight of its concept × share(t) × urgency (problems whose share(t) ≥ counted from)
urgency = 1 when it appears before the end of the plan, else max(floor, 1 − hours after the end / fading hours)
end of plan = the latest deadline of the objectives of the solution (now when they have none)
Foreseen = max over the dates judged of value(t) / threshold (1 at the threshold)
urgency = 1 when it appears before the end of the plan, else max(floor, 1 − hours after the end / fading hours)
end of plan = the latest deadline of the objectives of the solution (now when they have none)
Foreseen = max over the dates judged of value(t) / threshold (1 at the threshold)
Dates judged — in Scan mode (default) every date of the projections after now is judged and the worst one wins: the box tells when the worst moment comes. In Fixed mode only the chosen date is judged.
Contributions — each problem counted at the worst date gives a contribution FORESEEN on the field world: its concept, its name, the forecast that projects it, the date it appears, its share at the worst date and, when lower than 1, its urgency (“it appears after the end of the plan”).
04 · Parameters
In the supervision box
Defaults of getDefaultConfiguration, edited by the extension prospectiveDistanceStrategy.
| Parameter | Key | Default | Meaning |
|---|---|---|---|
| Judged at | mode | SCAN | Every date up to the horizon (worst point), or a fixed date (FIXED). |
| Fixed date | fixedDate | empty | Date (clock of the use case) judged in FIXED mode; empty: the horizon is scanned. |
| Counted from | minShare | 0.5 | Share of the forecast ensemble from which a projected problem counts. |
| Threshold | threshold | 0.3 | Weighted sum at which Foreseen reaches 1. |
| Actuality weight | weightActuality | 1 | The problem itself, the plan does not treat it. |
| Potentiality weight | weightPotentiality | 0.5 | A possible problem. |
| Risk weight | weightRisk | 0.3 | A risk the plan does not prevent. |
| Urgency fades over | urgencyHours | 24 h | Hours over which the urgency of a problem appearing after the end of the plan falls from 1 to the floor. |
| Urgency floor | urgencyFloor | 0.25 | Lowest urgency of a far problem; 0 ignores it. |
05 · Example
With the default configuration
An actuality foreseen by every curve of a forecast before the end of the plan, and treated by no objective, weighs 1 × 1 × 1 = 1: Foreseen = 1 / 0.3 ≈ 3.33, the alert box opens on its own. A risk certain on 60 % of the curves weighs 0.3 × 0.6 = 0.18, i.e. 0.6 of the threshold.
The same actuality foreseen 12 h after the latest deadline of the objectives has an urgency of 1 − 12 / 24 = 0.5 and counts for half; from 18 h after it the floor 0.25 applies.
06 · Watch points
Good to know
“Counted from” belongs to this strategy. The “Forecast certainty” share of the supervision box (timeline of r-iota and deduction with the forecasts) is a separate setting, 50 % by default too.
This page has no capture of the box: the supervision walkthrough was taken with the seven earlier detectors.
gind-detection-engine-prospective-distance-strategy/…/ProspectiveDistanceStrategy.javaProblems foreseen, shares, urgency, scan or fixed date, contributions.gind-npm-strategy-supervision-prospective-distance/src/prospective-distance-extension.tsThe parameters in the supervision box.gind-webapp-r-iota/…/supervision/SupervisionEngine.javaProjections of the forecasts put in the context, clock of the use case.gind-cep-forecast-projection/…/ForecastProjectionEngine.javaThe distributions (active, certain) over the curves of a forecast.