Hi All.
I’m running subject level models with a large number of volumes (>8,000 / subject) with a fast multi-band sequence (TR=0.41sec). Model estimation is completed for most subjects in 2 or 3 hours, but for some subjects it can take 24h. Convergence of non-spericity estimation (spm_est_non_sphericity (v6913) ) seems to very slow for these subjects and the memory usage builds up over time. The used memory can exceed 60 GB for a single model. Is this built-up of memory use to be expected? Are there some settings or tweaks that can cap or prevent memory usage?
I’m calling 'spm_get_defaults('stats.maxmem',2^34);’ before the model estimation to no apparent effect.
Below is the job configuration.
Thanks a lot.
Best,
Stephan
—
%-----------------------------------------------------------------------
% Job saved on 16-Apr-2018 09:10:53 by cfg_util (rev $Rev: 6942 $)
% spm SPM - SPM12 (7219)
% cfg_basicio BasicIO - Unknown
%-----------------------------------------------------------------------
matlabbatch{1}.cfg_basicio.file_dir.dir_ops.cfg_named_dir.name = 'subjfldir';
matlabbatch{1}.cfg_basicio.file_dir.dir_ops.cfg_named_dir.dirs = {'<UNDEFINED>'};
matlabbatch{2}.cfg_basicio.file_dir.dir_ops.cfg_cd.dir(1) = cfg_dep('Named Directory Selector: subjfldir(1)', substruct('.','val', '{}',{1}, '.','val', '{}',{1}, '.','val', '{}',{1}, '.','val', '{}',{1}), substruct('.','dirs', '{}',{1}));
matlabbatch{3}.spm.stats.fmri_spec.dir(1) = cfg_dep('Named Directory Selector: subjfldir(1)', substruct('.','val', '{}',{1}, '.','val', '{}',{1}, '.','val', '{}',{1}, '.','val', '{}',{1}), substruct('.','dirs', '{}',{1}));
matlabbatch{3}.spm.stats.fmri_spec.timing.units = 'secs';
matlabbatch{3}.spm.stats.fmri_spec.timing.RT = 0.41;
matlabbatch{3}.spm.stats.fmri_spec.timing.fmri_t = 16;
matlabbatch{3}.spm.stats.fmri_spec.timing.fmri_t0 = 8;
matlabbatch{3}.spm.stats.fmri_spec.sess.scans = '<UNDEFINED>';
matlabbatch{3}.spm.stats.fmri_spec.sess.cond = struct('name', {}, 'onset', {}, 'duration', {}, 'tmod', {}, 'pmod', {}, 'orth', {});
matlabbatch{3}.spm.stats.fmri_spec.sess.multi = '<UNDEFINED>';
matlabbatch{3}.spm.stats.fmri_spec.sess.regress = struct('name', {}, 'val', {});
matlabbatch{3}.spm.stats.fmri_spec.sess.multi_reg = '<UNDEFINED>';
matlabbatch{3}.spm.stats.fmri_spec.sess.hpf = 160;
matlabbatch{3}.spm.stats.fmri_spec.fact = struct('name', {}, 'levels', {});
matlabbatch{3}.spm.stats.fmri_spec.bases.hrf.derivs = [0 0];
matlabbatch{3}.spm.stats.fmri_spec.volt = 1;
matlabbatch{3}.spm.stats.fmri_spec.global = 'None';
matlabbatch{3}.spm.stats.fmri_spec.mthresh = -Inf;
matlabbatch{3}.spm.stats.fmri_spec.mask = {'/projects/stge3905/toolbox/spm12/tpm/mask_ICV.nii,1'};
matlabbatch{3}.spm.stats.fmri_spec.cvi = 'FAST';
matlabbatch{4}.spm.stats.fmri_est.spmmat(1) = cfg_dep('fMRI model specification: SPM.mat File', substruct('.','val', '{}',{3}, '.','val', '{}',{1}, '.','val', '{}',{1}), substruct('.','spmmat'));
matlabbatch{4}.spm.stats.fmri_est.write_residuals = 0;
matlabbatch{4}.spm.stats.fmri_est.method.Classical = 1;
|