On 7/24/06, Sebastian Haase <[EMAIL PROTECTED]> wrote:
How accurate do you want the distribution to be and what sort of offset is there? If the number of counts is greater that about 10 and you don't care too much about the poisson tail then a gaussian will work fine. If you have very large counts (>1e6), which I doubt, the accuracy of the poisson distribution becomes a tricky matter because it needs care to compute the factorial to the needed accuracy. The factorial turns up in the rejection methods used in the algorithms. Most of the algorithms also compute special constants that depend on the mean of the distribution, so while efficient for a fixed mean they are less so for varying means like you want.
I tend to just go with gaussian noise most of the time.
Hi,
Essentially I'm looking for the equivalent of what was in numarray:
from numarray import random_array
random_array.poisson(arr)
That is: if for example arr is a 256x256 array of positive integers, then this
returns a new array of random numbers than are drawn according to the poisson
statistics where arr's value at coordinate y,x determines the mean of the
poisson distribution used to generate a new value for y,x.
[[This is needed e.g. to simulate quantum noise in CCD images. Each pixel has
different amount of noise depending of what it's (noise-free) "input" value
was.]]
How accurate do you want the distribution to be and what sort of offset is there? If the number of counts is greater that about 10 and you don't care too much about the poisson tail then a gaussian will work fine. If you have very large counts (>1e6), which I doubt, the accuracy of the poisson distribution becomes a tricky matter because it needs care to compute the factorial to the needed accuracy. The factorial turns up in the rejection methods used in the algorithms. Most of the algorithms also compute special constants that depend on the mean of the distribution, so while efficient for a fixed mean they are less so for varying means like you want.
I tend to just go with gaussian noise most of the time.
Thanks,
Sebastian Haase
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Numpy-discussion mailing list
Numpy-discussion@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/numpy-discussion
------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________ Numpy-discussion mailing list Numpy-discussion@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/numpy-discussion
