Thanks - this would be a cleaner solution, however I already have one
set up using jobmanager and specify and estimate model. What would I
gain by doing this using my own code? I guess smoothness estimation
would not be run and save some time. What percentage of the
computation (more or less?) done by estimate is smoothness estimation?
I still think "save residuals" would be a useful option :)
Thanks!
Chris
On Fri, Aug 6, 2010 at 10:28 AM, Guillaume Flandin
<[log in to unmask]> wrote:
> Hi Chris,
>
> the data type of the temporary standardised residual images is defined in
> spm_spm.m line 533.
> If you want SPM to save all the residual images you need to edit
> spm_defaults.m so that stats.maxres = Inf.
> These images are then deleted (as they are only used for estimating the
> smoothness) in spm_spm.m lines 912-918.
> Note also that they are not the raw residuals but standardised ones (see
> line 764 of spm_spm.m) as it's what spm_est_smoothness.m later expects.
>
> However, if you just want to remove some confounds in your data, you might
> be better off at writing some dedicated code that would call
> x = spm_sp('Set',X);
> r = spm_sp('res',x,Y);
> where X is your design matrix containing the confounds, Y your time series
> and r the residuals.
> Using spm_spm.m to only get the residuals seems to me more of a workaround
> than a feature this function should have.
>
> Best regards,
> Guillaume.
>
>
> On 06/08/10 15:09, Chris Filo Gorgolewski wrote:
>>
>> Thanks all.
>>
>> On Fri, Aug 6, 2010 at 9:42 AM, Stephen J. Fromm<[log in to unmask]>
>> wrote:
>>>
>>> Why do you think using double instead of float is that much more time
>>> consuming? (I do understand that a naive calculation says it takes twice as
>>> much time to write/read the data.)
>>
>> The naive calculations has proven to be correct :)
>>
>>> If you found the place in the code to turn off the deletion of the
>>> residuals, maybe you could find the place in the code where SPM uses double
>>> and change it to float?
>>
>> I ended up converting to float32 using fslmath. I prefer to change SPM
>> code as little as possible (because it is hard to maintain your own
>> version).
>>
>> BTW - can I make a feature request? Could saving residuals be an
>> option I could set somewhere? This way I would not have to patch my
>> SPM version each time there is a new release.
>>
>> Thanks!
>> Chris
>>
>>
>
>
> --
> Guillaume Flandin, PhD
> Wellcome Trust Centre for Neuroimaging
> University College London
> 12 Queen Square
> London WC1N 3BG
>
>
|