| 2D K-S test [message #56519] |
Thu, 25 October 2007 09:49  |
ianpaul.freeley
Messages: 18
Registered: March 2007
|
Junior Member |
|
|
Can someone point me to some IDL code that can do a 2D Kolmogorov-
Smirnov Test? I'm feeling lazy and figure someone else must have
ported it from Numerical Recipes by now.
cheers,
I.P. Freeley
|
|
|
|
| Re: 2D K-S test [message #56570 is a reply to message #56519] |
Sat, 27 October 2007 08:11  |
ianpaul.freeley
Messages: 18
Registered: March 2007
|
Junior Member |
|
|
Thanks Wayne. I went and coded it up and wondered about that issue but
didn't do anything about it. I'll go back and add some if statements.
-IP
On Oct 26, 12:46 pm, wlandsman <wlands...@gmail.com> wrote:
> On Oct 26, 9:29 am, Loren Anderson <ander...@bu.edu> wrote:
>
>> Hi, I just spent a while working with the K-S test, but I'm confused
>> as to what you are trying to do. The K-S test seems to me to be one-
>> dimensional by nature. Can you clarify a bit?
>
> This is discussed in the section of Numerical Recipes titled "Do Two-
> Dimensional Distributions Differ" (Section 14.7 in my edition.)
> The idea is to determine if a 2-d distribution of points is consistent
> with another 2-d distribution (or a model). You are right that one
> cannot directly generalize the 1-d K-S test because a cumulative
> probability distribution is only defined in 1d, so one needs to define
> a surrogate.
>
> Back to the OP's question. I am not aware of any IDL
> implementations of the code. (I started but never finished working
> on it.) If you do code it yourself and have an old edition of NR,
> check out the bug fix reported athttp://www.nr.com/forum/showthread.php?t=576
>
> --Wayne
|
|
|
|
| Re: 2D K-S test [message #56575 is a reply to message #56519] |
Fri, 26 October 2007 10:46 |
wlandsman
Messages: 743
Registered: June 2000
|
Senior Member |
|
|
On Oct 26, 9:29 am, Loren Anderson <ander...@bu.edu> wrote:
> Hi, I just spent a while working with the K-S test, but I'm confused
> as to what you are trying to do. The K-S test seems to me to be one-
> dimensional by nature. Can you clarify a bit?
This is discussed in the section of Numerical Recipes titled "Do Two-
Dimensional Distributions Differ" (Section 14.7 in my edition.)
The idea is to determine if a 2-d distribution of points is consistent
with another 2-d distribution (or a model). You are right that one
cannot directly generalize the 1-d K-S test because a cumulative
probability distribution is only defined in 1d, so one needs to define
a surrogate.
Back to the OP's question. I am not aware of any IDL
implementations of the code. (I started but never finished working
on it.) If you do code it yourself and have an old edition of NR,
check out the bug fix reported at
http://www.nr.com/forum/showthread.php?t=576
--Wayne
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
