|
From: Daniel J S. <dan...@ie...> - 2004-03-05 16:40:24
|
Justace Clutter wrote: >Cool, > > My stuff is almost working and with the image patch much faster. The >new deal is this. Since I am generating a correlation matrix with >respect to one set of channels, i.e. the channels correlated with >themselves, half of the matrix is useless. Specifically the upper left >of the diagonal. So, to cut back on my calculation time I fill that >region with NaN, but how do I put that in gnuplot so that it does not >plot it? I was putting it in as a -2 and then telling it to only plot >[-1:1] on the z but that was silly since it is a 2D plot. Maybe the >image style needs to have an internal variable that the user can set to >define NaN's. That way I could go and set that to -2 and when image >converts the float to a color it goes ohhh this is a NaN, what fill >style should I use here or what predefined color do I have for NaN. >Just some thoughts and questions here. > For your correlation matrix, you might want to simply replicate the other half since it is probably Hermitian symmetric, of if you are working with real signals, outright symmetric. In any case, your idea of mapping to a certain value might work. It may cause a problem with automatic scaling of the color axis. However, you should be able to manually set the color axis so that your "NaN" number gets mapped to white. For example, set the color map some how to [-1:1] and then make your "NaN" be a +2 (Or -2, whichever gets mapped to the white value.) Dan |