?

Log in

No account? Create an account
Project

Bah!

I finally get my code working, and then run a diff between the output of a known "working" version, and the output of my code. I get about 30 lines that show up as different.

< 426 8.05078e-09
---
> 426 8.05077e-09

Now, the trouble is that I'm working with floats. Given that up until one of the very last iterations all the variables match, I've a feeling that the difference may be that of the rounding errors from a straight sum of a matrix, as apposed to splitting the matrix, then summing the sums of the split matrices. But I have no way of knowing :(

Will speak to my project supervisor on Monday...

Comments

Not that it helps much but it smells suspiciously like rounding errors to me. Hows things going more generally with project now? Oh and are you coming to entropy tonight?
I think they are rounding errors, especially since it's the least sig fig that's varying. Amusingly my code may actually be more accurate, since it adds small numbers to small numbers then adds the resultant bigger numbers, what an amusing side effect of parallelisation :)

Oh, Entropy, I've not been for aaages... Quite possibly.
An error that small is pretty musch inevitable while using floats.

There's a paper 'what every computer scientist should know about floating point' which you should probably read.
I was thinking this might be the case, I'll look up that paper...
fireworks

August 2010

S M T W T F S
1234567
891011121314
15161718192021
22232425262728
293031    
Powered by LiveJournal.com