HeuristicsTable of content
The Feedback heuristic is very different from other heuristic nodes. It uses validated path within the scope of a single pathfinding node to make validated traversal more or less likely to be traversed by subsequent queries. There is two main “modes” : local & global feedback.
Properties
| Property | Description |
|---|---|
| Basics | |
| Weight Factor | Weight of this heuristic against other concurrent heuristics. The higher the value, the more important it is during resolution. |
| Invert | Whether the score of this heuristic should be inverted. This effectively samples the score curve backwards. |
| Score Curve | Curve over which the heuristic values will be remapped. |
| Feedback Settings | |
| Visited Point Weight Factor | Multiplier to be applied to traversed nodes. |
| Visited Edge Weight Factor | Multiplier to be applied to traversed edges. |
| Global Feedback | Whether this feedback should be global or local |
| Affect All Connected Edges | If enabled, the extra score affect all edges connected to the traversed Vtx, not just the one actually traversed. |
| Local Weight | |
| Use Local Weight Multiplier | If enabled, this heuristic will be using a dynamic, per-point weight factor. |
| Local Weight Multiplier Source | Whether to read the weight from Vtx or Edges points. |
| Local Weight Multiplier Attribute | Attribute to read the local weight from. |
| Roaming | |
| UVW Seed | Bounds-relative roaming seed point |
| UVW Goal | Bounds-relative roaming goal point |
Roaming seed/goal points are used as fallback in contexts that are using heuristics but don’t have explicit seed/goals; such as Cluster Refine‘MST or Score-based refinements.
Global vs Local feedback
Global feedback is applied in the scope of the entire node: each completed query will score down the next ones.
As soon as a query is resolved (a Seed is connected to a Goal), the Vtx and Edges it traverses are scored by their specified factors.
If global feedback is disabled, the extra scoring is only processed for the duration of the query – as such it only works with plots.
In order to be properly accounted for, global feedback requires queries to be processed one after another, greatly affecting processing time!