 |
 |
Dear Qinghua, > OK, i got the idea that if the registration picture looks fine, i > shouldn't need to worry a bout that warning. > but i am really curious if it's true that in version 4.1.4 they just > hide those warnings, and in 4.1.5 they just show that? that make > sense when i was using the same design (no change at all), but get > different log files in those warnings, but all others are the same. the messages have never been hidden (though I probably will hide them in the future since they cause people to much distress). There may be several reasons why you see them now on data where previously you didn't. There are constantly little tweaks and fixes done to fnirt, so it wouldn't be identical in both releases. I sometimes see these messages when fnirt is "too close" to the "solution" for some level. The derivatives have pretty much disappeared. That means that a change that helps fnirt converge in fewer iterations (I know I have made a few of those) could potentially lead to this warning being printed for the last iteration/iterations. The important thing to know is that fnirt never "ends" on one of these bad notes. It always adjusts the matrix until the problem is sufficiently well conditioned before it takes a step. Hope that helps. Jesper > just want to make sure. > thanks, > -Qinghua > > Date: Fri, 23 Apr 2010 21:05:59 +0200 > From: [log in to unmask] > Subject: Re: [FSL] Feat Registration Question > To: [log in to unmask] > > Dear Qinghua, > > > I have a question regarding the registration part in Feat. > previously, when my FSL version was 4.1.4, the registration runs > smoothly (without warning or error). > > but, after i installed new version of FSL (4.1.5), i found that the > registration for my new subject got some warnings. and then, i just > re-run the same subject i have already done to check out if this is > because of the subject, but again i got these warnings, quote from > the log file: > "Setting subsampling > Setting reg mode > Setting lambda > SpMat::SolveForx: Warning requested tolerence not obtained. > Requested tolerance was 0.001, and achieved tolerance was 0.851883 > This may or may not be a problem in your application, but you should > look into it > SpMat::SolveForx: Warning requested tolerence not obtained. > Requested tolerance was 0.001, and achieved tolerance w! > as 0.160804 > This may or may not be a problem in your application, b > ut you should look in! > to it > this is an error message from fnirt. For every iteration fnirt > solves a large system of equations numerically. That solution can be > more or less stable (the coefficient matrix can be close to > singular). When it encounters a close to singular coefficient matrix > it prints these error messages. > > However, there is an inherent "stabiliser" built into fnirt so that > for the next iteration it will make sure the matrix is better > conditioned. Hence these warnings are usually not a problem when you > just see a couple of them and the registered images still look good. > > I cannot say why it appears now on the same data where before it > didn't, but even very small changes in the code (and subsequently in > this matrix) can have effect on the conditioning of the matrix. > > In conclusion, if the registered images look good this is nothing to > worry about. > ! > > Good Luck Jesper > > > Hotmail: Trusted email with powerful SPAM protection. Sign up now.