Research Plan
Thoughts
- We will rerun Heuristic Comparison Run
- We will rerun Greedy Random Analysis Run
- We are considering 100 customer problems
- Our goal is have all experiments done by November 14th
- We want to log GR specially (already somewhat implemented)
- We would like to investigate a Steepest Climbing GR look-a-like
- We want to write a StatAnalyzer that lets us export data for Statistical Analysis (and also measures from the best known solutions)
- As for the paper, we'd like to write it in 'hypothesis-test-conclusion' form
- We might want to run a GR change heuristic, which compares large changes to small changes. Change is defined as the of the sum of the absolute values of the deltas of the truck routes after a move.
- We'd like to measure as much as possible so that we can formulate hypothesises and correlations between a certain characteristic of a move and the quality of the result
- We might want to examine a one-ply vs. GR change of different solution ranks
- We will run a test where we leave the moved customer at high (rather than medium)
- We might want to run a test where we leave the moved customer at low (rather than medium)
- We may want to do randomly set a customer on each route to medium... does RPGR already do this to a certain extent? It isn't particularly successful.
- Something that we might want to look at is GR's improvement variation... e.g. 10 cycles on the same solutions, and seeing how the end solutions vary.
Dreams
Ideas that would be cool, but might be constrained by time:
- "Session Analyzer" - A tool to recreate the moves and stages that optimized a solution. This might work by parsing through the log files and then creating scripts that are somehow "run" by another program (maybe a web script, or a standalone app). Essentially, "Session Analyzer" would let us watch the sessions happen.