Closed bgottula closed 3 days ago
I root caused the bad behavior to the dtype
of the distance_matrix
Numpy array in tsp.solve_route()
. The dtype
of the array was left to the default which is a floating point type, but Google's OR-Tools expects integer distances. Apparently some older version of OR-Tools was able to coerce the floats to integers, but now passing a floating point distance (even if the value stored in the float is an integer) results in solutions that are clearly suboptimal.
align
is supposed to visit a set of locations on the sky in a (nearly) optimum order, using a travelling salesman solver. I just ranalign
after some recent upgrades and it very clearly was not following the optimum path. It performed several meridian flips when only one meridian crossing should be necessary.Related to this: Right now I'm using ortools or something from Google, however I recently learned that the
networkx
package may have similar functionality in case I need a replacement.