a note on my previous question
jcrofts
Posted on 07/29/08 08:15:06
Number of posts: 29
j crofts
I should add that depending upon my choice of seed I do not always get the matlab warning.
Untitled Post
rmcintosh
Posted on 07/29/08 09:28:26
Number of posts: 394
Hi Jonathan,
Perhaps you could give me a bit more detail about what sort of data you are analysing? Also I would suggest checking the data in the seed voxel just to see what sort of values you are getting. Its possible that there may be NaNs or complex numbers in the data, though I can't really understand why that would be.
cheers
Randy
Untitled Post
jcrofts
Posted on 07/29/08 10:25:04
Number of posts: 29
j crofts
Hi Randy,
The data I am looking at is from a reversal learning and attentional set shifting experiment in rodents. I have four conditions; 12 subjects in each; and 36 brain regions are included. We have data for the 2DG uptake in each brain region over the 4 different conditions. I hope this is useful, I am a mathematician by trade and am not very up to date with some of the neuroscience terminology.
I have had a closer look at the m-file, pls_analysis.m (05-JAN-2005 by Jimmy Shen), the error is occurring on line 1387, u_se = sqrt((u_sq-u_sum2)/(num_boot)), for the seed, (u_sq-u_sum2) should be zero, I think, but it actually has the value -0.0 in matlab causing the error. Should I just replace (u_sq-u_sum2) by abs(u_sq-u_sum2) in pls_analysis.m?
Jonathan
Untitled Post
rmcintosh
Posted on 07/29/08 10:52:54
Number of posts: 394
Hi Jonathan,
The fix you propose should work. I did have a check for this sort of thing in early versions of the PLS code. This seems to be a precision issue in MATLAB, so although its zero with rounding, at the 60th decimal point its negative.
Randy
Untitled Post
I'm Online
jshen
Posted on 07/29/08 10:59:45
Number of posts: 291
Thanks for the feedback from Randy and Jonathan, I will apply this fix in the next version of PLSgui and PLScmd (pls_analysis.m) release.