Other Pinewood Derby Race Methods
Responses to my Challenge:
Entries:
- Algorithm described at the end of Darin McGrew's Race Methods document.
- Algorithm and program (get demo copy) by
which imputes relationships between cars by applying transitivity: "If a heat shows that car A is probably faster than car B and another heat shows that car B is probably faster than car C, then we may imply that car A is probably faster than car C."
The algorithm is in the "elimination" family. The elimination criterion is parametric.
Once cars are inserted into the initial grid, the pairings are deterministic, i.e. have no random factors. (This satisfies my auditability requirement.)
Clark has submitted the following formal evaluation results. K is a "lane difference" measure.
60 races, 28 cars, 3 lanes
|
K | 0 | 1 | 2 | 3 | 4
|
"4 Trophy" Accuracy | 100% | 74% | 62% | 51% | 45%
|
"Top-4" Accuracy | 100% | 98% | 96% | 94% | 90%
|
100 races, 28 cars, 3 lanes
|
K | 0 | 1 | 2 | 3 | 4
|
"4 Trophy" Accuracy | 100% | 85% | 80% | 77% | 73%
|
"Top-4" Accuracy | 100% | 99% | 98% | 98% | 97%
|
Back to the Evaluation Method page...
Back to the Stearns Evaluation page...
Latest update: 9/27/98
Copyright 1997, 1998 © by Stan Pope. All rights reserved.