Thread View: sci.math.stat
2 messages
2 total messages
Started by Brian Marick
Thu, 16 Apr 1992 22:11
Smoothing
Author: Brian Marick
Date: Thu, 16 Apr 1992 22:11
Date: Thu, 16 Apr 1992 22:11
35 lines
1950 bytes
1950 bytes
I am looking for references that might help me with the following problem. (In short, I need references to data smoothing techniques.) For the UNIX kernel, I can measure the effects of two input distributions, but I cannot capture them directly. To simplify, let's say that the effect is the number of times each routine in the kernel (total around 5000) is called. I can linearize this data by numbering routines in the order they're found in the source. Related routines will be close to each other (to a *very* rough approximation -- but probably good enough for now). I can use the routine number as the X axis, entry count as the Y axis. This will be a wild point blur: calls to adjacent routines may differ by a factor of more than a million. I need to compare two point blurs to see (very roughly) whether they're similar or quite different. What I am thinking of doing is smoothing them to a curve and matching the curves by eye. I know about Tukey/EDA style smoothing. I know a little bit about LOWESS. I don't know enough to say whether or when each is appropriate. (That is, I need to read papers or books that say more than just "this is a technique"; I also need "... and here's where it's most useful, and here are the dangers".) Where can I learn what I need? Also, I need programs to support whatever techniques I use. I have ready access to SYSTAT and STATA. A completely different approach would be fine, too. I chose this because it's visual and intuitive. Thanks. Brian Marick, marick@cs.uiuc.edu, uiucdcs!marick Testing Foundations: Consulting, Training, Tools.
Re: Smoothing
Author: dmatthews
Date: Fri, 17 Apr 1992 12:55
Date: Fri, 17 Apr 1992 12:55
20 lines
1457 bytes
1457 bytes
In article <1992Apr16.221121.20217@m.cs.uiuc.edu>, marick@m.cs.uiuc.edu (Brian Marick) writes: > >I am looking for references that might help me with the following >problem. (In short, I need references to data smoothing techniques.) > ... >A completely different approach would be fine, too. I chose this >because it's visual and intuitive. Why not FFT the data, MULTIPLY it by a suitable transfer function (lowpass filter) which you draw yourself, and inverse FFT the multiplied data? You can get the impulse response by inverse FFTing the transfer function. Avoid steep sides and sharp corners in your lowpass filter. This way you know exactly what you're doing to the data. The transfer function must multiply both the real and imaginary parts of the FFT. Email me with further questions . Easy -- use available FFT package, write the filter in a few lines. Dr. David L. Matthews, IPST, Univ. of Maryland, College Park MD 20742-2431 Telephone (301)405-4830 Internet dmatthews@uap.umd.edu FAX (301)314-9363 NSI/SPAN UMDUAP::DMATTHEWS
Thread Navigation
This is a paginated view of messages in the thread with full content displayed inline.
Messages are displayed in chronological order, with the original post highlighted in green.
Use pagination controls to navigate through all messages in large threads.
Back to All Threads