 |
| I probably should have had more than this today. |
Clicking on this item from my RSS (side note, the correct answer is yaki-gyoza, obviously) started me thinking about the following problem. Given a poll, what's the best way to estimate the minimum number of poll responses necessary to create the observed result distribution? My initial thought is that you take the minimum response separation. When I clicked, this poll had yaki: 70.79%; omnomnom: 14.85%; wrong: 14.36%. The minimum separation is 0.49%. Assume this is one person. Then the results have [144.469 30.306 29.306]. Hrm. Big fractional residuals. Two people? [288.939 60.612 58.612]. Still pretty bad. Three? [433.408 90.918 87.918]. And then I wrote a perl script, and it looks like dN = 20 is the lowest solution, although the residual is still pretty large (0.63). dN=7 has lower residual, but doesn't round to the right solution. This seems to suggest that the residual isn't the best metric here. There's a periodicity in the residual, so that should limit the number of points that need to be checked, but it still seems like there's no great algorithm other than "check a bunch of possibilities."
- Dog.
- Cat and owl.
- Beep beep! Space shuttle coming through!
- At least Pikachu is trying.
- Marnie isn't playing here anymore.
No comments:
Post a Comment