Vrpy not accomplish the Time Windows Constraints

I am trying to solve a CVRPTW on a graph with 54 nodes and 353 edges, as you can see it is a little size problem, and all the edges have a time window, a demand and a service time as the next image:

However, when I try to solve the problem with the exact algorithm the best routes are not satisfying the time windows constraint, as an example (even when this happens with many nodes) the previous node 36 has the time window [8 , 14] and when I look at the prob.arrival_time attribute the arrival time at the node is 15.3 (out of the TW) as shown in the next image

You can reproduce the problem with the next graph in the zip and this code:

from networkx import read_gpickle
import vrpy
G = read_gpickle('graph_vrpy,test')
len(G.nodes)
prob = vrpy.VehicleRoutingProblem(G, load_capacity=80)
#prob = VehicleRoutingProblem(G, load_capacity=100)
prob.time_windows = True
prob.solve(exact=True)

prob.best_routes
prob.arrival_time
G.nodes['36']

graph_vrpy.zip

[torressa]

And… Going back to the beginning, @tomatoes-prog is correct. With the previous arrival_time version, without the min, cspy works with direction="forward" and not with "both". Performance-wise "forward" seems to be faster so it’s not really a problem, I just need to double check the backward REF.

PS. Sorry for committing to master

[torressa]

Can confirm, only for arrival times though… I’ve just added another condition there