openego / eDisGo / 35856695870

84%

push

github

fix(tools): do not abort pseudo coordinates on a meshed LV grid

make_pseudo_coordinates aborted with

    NetworkXError: The node BranchTee_mvgd_32064_lvgd_8829100000_5
    is not in the graph.

on MV grid 32064, and with it the whole spatial_complexity_reduction, which
generates pseudo coordinates first (apply_pseudo_coordinates, default True).

_make_coordinates lays the buses out breadth-first from the station and
assumes a tree, where every node is reached from exactly one neighbour. The
LV grid in question is not one: LVGrid_8829100000 holds 41 buses on 41 lines,
with a ring of three branch tees. A node on a ring is discovered from both
sides and was appended to the queue twice; the node is removed from the
working copy when its first entry is processed, so the second entry called
nx.neighbors on a node that was no longer there. Reproduced on the grid and
reduced to a six-node graph: a ring plus a tail, the tail being what keeps the
loop running long enough to reach the duplicate.

A node is now queued at most once and keeps the coordinates of its first
discovery, which is the one the breadth-first order reaches over the shorter
path. The neighbour counter still counts skipped neighbours, because it is the
index of the branch in the fan-out and has to stay in step with the neighbour
count handed to coordinate_branch. In a tree nothing is ever skipped, so
tree-shaped grids keep their coordinates exactly -- verified by replaying a
tree against both implementations and pinning the result in a test.

The loop also no longer runs the queue dry on a graph that is not fully
connected: leftover buses keep their original coordinates and are named in a
warning instead of raising IndexError.

Grid 32064 now completes the spatial reduction (7679 -> 2493 buses).

openego/eGo#227

6 of 7 new or added lines in 1 file covered. (85.71%)

9556 of 11378 relevant lines covered (83.99%)

0.84 hits per line