The Routing Apprentice

Four trucks, thirty customers, one depot (blue square). A 5k-parameter policy network learns the skill of routing by building millions of routes on RANDOM maps and being rewarded for shorter total distance. The maps below are all the SAME fixed test map — one it never trains on. Benchmarks on this map: nearest-neighbour heuristic 6.97, with 2-opt polishing 6.53. Act IV-B: a professional skill, learned from the rules alone.

The same map, as the apprentice learns

epoch 0

total distance 7.06

epoch 3

total distance 7.16

epoch 8

total distance 7.14

epoch 15

total distance 7.28

epoch 30

total distance 6.81

epoch 60

total distance 6.65

epoch 100

total distance 6.66

epoch 150

total distance 6.86

epoch 225

total distance 6.68

epoch 325

total distance 6.68

epoch 450

total distance 6.68

epoch 600

total distance 6.59

epoch 800

total distance 6.47

epoch 1000

total distance 6.48

Distance on the test map over training

epochdistancevs 2-opt benchmark
07.056+8.0%
37.165+9.7%
87.141+9.3%
157.281+11.5%
306.807+4.2%
606.648+1.8%
1006.662+2.0%
1506.857+5.0%
2256.684+2.3%
3256.684+2.3%
4506.684+2.3%
6006.591+0.9%
8006.466-1.0%
10006.484-0.7%