error using 3dLSS

AFNI Message Board
1

Dear AFNI Experts,

In trying to get trial-specific beta-weight estimates to eventually use with MVPA analyses, I’m using 3dDeconvolve with its “-x1D” option to save out a regression matrix, which I then input to 3dLSS. My individual stimulus regressor ends up being in column 7, and when I run 3dLSS on a single stimulus (having modeled all other stimuli separately in 3dDeconvolve), I get this error:
** FATAL ERROR: Matrix attribute BasisColumns_000007 has illegal value

That column is the same length as the others (equal to the total TRs), and does have some non-zero values (though only a few because it’s for just a single stimulus). I tried updating AFNI to its current version as of today, but get the same error. I’d be happy to provide any other info that might help you help me!

-Will

2

Just to briefly follow up with this, it still has me stumped. I double checked, and all the columns are indeed the same length (corresponding to 625 total TRs). I would be happy to upload the regression matrix from 3dDconvolve if that would help. In running 3dDeconvolve, I only have one -stim_times_IM regressor, and all the stim_times regressors use the “BLOCK(0.5,1)” style HRF. So 3dDeconvolve runs without error, but 3dLSS simply will not stop complaining about the regression matrix I’m giving it. I’d really appreciate any thoughts or suggestions.

Best,
Will

3

Email me the matrix file and the exact 3dDeconvolve command you used to create the matrix. I’ll take a look at that.

robertcox at mail.nih.gov

4

Since Bob helped me with this by individual email, I thought I’d quickly post the solution, in the unlikely event anyone else runs into the same problem.
The error came from my having only one stimulus event in my regression matrix file. The old way of getting individual stimulus estimates, using only 3dDeconvolve, required modeling one stimulus event at a time, so I was just assuming the same thing would work with 3dLSS. But instead, 3dLSS requires multiple stimulus events, so that each one can be modeled separately from the others.

And now it’s working great!

Thanks,
Will