Processing math: 100%

Supported versions: latest (3.8) 3.7 3.6 3.5 3.4 3.3 3.2 3.1 3.0 main dev
Unsupported versions:2.6 2.5 2.4

Cost Matrix - Category

proposed

Warning

Proposed functions for next mayor release.

  • They are not officially in the current release.
  • They will likely officially be part of the next mayor release:
    • The functions make use of ANY-INTEGER and ANY-NUMERICAL
    • Name might not change. (But still can)
    • Signature might not change. (But still can)
    • Functionality might not change. (But still can)
    • pgTap tests have being done. But might need more.
    • Documentation might need refinement.

General Information

Synopsis

Traveling Sales Person - Family of functions needs as input a symmetric cost matrix and no edge (u, v) must value .

This collection of functions will return a cost matrix in form of a table.

Characteristics

The main Characteristics are:

  • Can be used as input to pgr_TSP.
    • directly:when the resulting matrix is symmetric and there is no value.
    • It will be the users responsibility to make the matrix symmetric.
      • By using geometric or harmonic average of the non symmetric values.
      • By using max or min the non symmetric values.
      • By setting the upper triangle to be the mirror image of the lower triangle.
      • By setting the lower triangle to be the mirror image of the upper triangle.
    • It is also the users responsibility to fix an value.
  • Each function works as part of the family it belongs to.
  • It does not return a path.
  • Returns the sum of the costs of the shortest path for pair combination of nodes in the graph.
  • Process is done only on edges with positive costs.
  • Values are returned when there is a path.
    • The returned values are in the form of a set of (start_vid, end_vid, agg_cost).
    • When the starting vertex and ending vertex are the same, there is no path.
      • The agg_cost int the non included values (v, v) is 0.
    • When the starting vertex and ending vertex are the different and there is no path.
      • The agg_cost in the non included values (u, v) is .
  • Let be the case the values returned are stored in a table, so the unique index would be the pair: (start_vid, end_vid).
  • Depending on the function and its parameters, the results can be symmetric.
    • The agg_cost of (u, v) is the same as for (v, u).
  • Any duplicated value in the start_vids are ignored.
  • The returned values are ordered:
    • start_vid ascending
    • end_vid ascending
  • Running time: approximately O(|start_vids|(VlogV+E))