Darwinbots Forum

Code center => Bugs and fixes => Bug reports => Topic started by: Sprotiel on December 19, 2006, 08:49:56 PM

Title: Bug in statistics
Post by: Sprotiel on December 19, 2006, 08:49:56 PM
I think every per species average wrongly divides the sum of the relevant properties by the total number of bots instead of dividing by the number of bots of the species. E.g. if you start a sim with 10 vegs and 5 animals with 5000 nrg each, the average energy for vegs will appear as 3333, and for animals as 1667.
Title: Bug in statistics
Post by: EricL on December 20, 2006, 12:08:06 PM
You are exactly right.  Nice find.  Fixed in 2.42.9q.
Title: Bug in statistics
Post by: Sprotiel on January 21, 2007, 03:10:33 PM
It seems my former bug report was incorrect. In 2.42.9t (and probably also in .9u), mutations and mutations/DNA length appear to be incorrectly multiplied by a factor (total number of bots / number of bots in species).
Title: Bug in statistics
Post by: EricL on January 25, 2007, 10:42:18 AM
Quote from: Sprotiel
It seems my former bug report was incorrect. In 2.42.9t (and probably also in .9u), mutations and mutations/DNA length appear to be incorrectly multiplied by a factor (total number of bots / number of bots in species).
I've been looking at this and I don't think this is the case.

Mutations is the average # of mutations per individual for that species so it is calculated as the sum of the all mutations for all members of the speces divided by the number of individuals in the species.

Mutations/DNA Length is the same just divided by dna length I.e.  Sum of (number of mutations for bot / DNA lenght of bot) for all bots in species.

I may be missing somehting, but I think it's correct.
Title: Bug in statistics
Post by: Sprotiel on January 25, 2007, 11:17:08 AM
Quote from: EricL
I may be missing somehting, but I think it's correct.
It probably is. I've seen since that I had some bots with a very large number of mutations. So the problem is just that the average is sometimes misleading.
Title: Bug in statistics
Post by: Numsgil on January 25, 2007, 11:11:59 PM
Yeah, I think using averages isn't the best idea.  Quartile plots vs. time would probably be better.  It would let you see both the average and the spread.
Title: Bug in statistics
Post by: EricL on January 25, 2007, 11:57:59 PM
And historgrams.  I'd love to see the distribution of offspring, nrg levels, percentage of coding DNA, etc....  I have a whole list of contained little projects like that someone with spare time could work on....