Experiment Ideas
Heuristic Comparison Run II (HCR2)
This is just a re-run the original Heuristic Comparision Run.
Greedy Random Analysis Run II (GRAR2)
This is just a re-run the original Greedy Random Analysis Run.
Multi-Move Greedy Random (MMGR)
This run compares greedy randoms making 1, 2, 5, 10 and 25 initial moves. We are looking for the breaking point, where greedy random stops finding improvements. Once we find it, we can zoom in on it and re-run the run with a small window to isolate the breakpoint. We can then try figure out why the breakpoint occurs there.
Steepest Climbing Greedy Random vs. Greedy Random Comparison Run (SCCR)
We compare the performance of the Steepest Climbing Greedy Random Lookalike heuristic versus the performance of Greedy Random. If SCCR preforms better (as I for some reason think it will), then we will want to develop further experiments to determine why... hopefully we will have proof from our GR experiments that affirms why SC is better.
GR Change Analysis Run (GRCAR)
By change we mean the sum of the absolute values of the deltas of the truck routes after a move. In GRCAR, we compare GRs that are constrained to making various levels (high, medium, low) of change. Before running this experiment, we make sure that we log "change" somehow and see what the range and average of change is... it may even be unnecessary because we will see a correlation from our data.
GR Non-constrained vs. GR (NCCR)
Non-constrained GR leaves the moved customer at high. Essentially we want to see if it is any better, worse or about the same (most likely "nothing" because the customer will hop right back to its old route, defeating the purpose of GR. Or at least that's what we think...). A collary of this is to set the moved customer to low and see what happens (most likely "bad things" because there's no way to kill any infeasibility on that route).
Greedy Random Improvement Variation (GRIV)
This experiment would run GR several times on the same solution and then analyze the variation of the solutions. Ideally, we would like to see no variation. An interesting parameter we might want to research is whether there's a correspondence between the number of cycles and GR's variance...