Hi Andrew,

There was something wrong with your .tgz file, though I was able
to extract the X-matrix from it.

By the way, the timing_tool.py program is currently setup to deal
with local times, only. But your times are global. Note that 298
seconds is your run time (a single run).

In any case, I think the early assessment was correct, in that some
subset of your regressors is not linearly independent. With 481
regressors in your model, which uses AM deconvolution, it isn't a
bit surprise.

I must say, using AM (or AM2) with TENT functions seems dangerous.
That gives you a lot of regressors per stimulus class, and this
sort of problem is common under such conditions. Even if you did
not run into multi-collinearity, the results would have to be
extremely noisy and unreliable. Small effects from noise can have
a large effect on the results.

Do you need to do deconvolution here? Is there any chance that a
fixed-shape response might be appropriate?

- rick