Thanks Michael! Your suggestion was quite useful. I successfully loaded the connectivity matrix into Matlab using ciftiopen and did the operation that I wanted (i.e. selecting part of the matrix).
Reza On Thu, Oct 25, 2018 at 6:41 PM Harms, Michael <mha...@wustl.edu> wrote: > > > As an aside, if ‘ft_read_cifti’ is expanding all the data to type double > (64 bit), then you are looking at needing > 66 GB to load a 91282x91282 > dconn.nii in matlab. I assume that is what is happening based on your > experience. > > > > If you want to see how far you can get in Matlab, I would at least try > switching to using ‘ciftiopen.m’ to load it into matlab, which uses the > gifti toolbox, which preserves the underlying data type of the .dconn.nii > (type single; i.e., 32 bit) upon load into matlab. > > > > I tested that just now on a system with 64 GB of RAM (and 64 GB of swap), > and after what must have been at least 5 min, it loaded successfully into > matlab (R2015a) with a size in matlab of 32 GB (and a type of ‘single’). > Note however that as that process was ongoing, the memory usage of the > overall system hit 56 GB (not counting cached memory), which decreased to > 33 GB when the load was completed (presumably that is reflective of there > essentially being two copies in memory during the loading process). > > > > Working with a dconn in matlab is inevitably going to be cumbersome and > slow, but by using this approach you might actually be able to get it > loaded and see where you can get from that point. Obviously, you would > have to be extremely conscious of memory issues for whatever operations you > try to perform in matlab. > > > > Cheers, > > -MH > > > > -- > > Michael Harms, Ph.D. > > ----------------------------------------------------------- > > Associate Professor of Psychiatry > > Washington University School of Medicine > > Department of Psychiatry, Box 8134 > > 660 South Euclid Ave > <https://maps.google.com/?q=660+South+Euclid+Ave&entry=gmail&source=g>. > Tel: 314-747-6173 > > St. Louis, MO 63110 Email: mha...@wustl.edu > > > > *From: *<hcp-users-boun...@humanconnectome.org> on behalf of Reza > Rajimehr <rajim...@gmail.com> > *Date: *Thursday, October 25, 2018 at 3:17 AM > *To: *"Glasser, Matthew" <glass...@wustl.edu> > > > *Cc: *"hcp-users@humanconnectome.org" <hcp-users@humanconnectome.org> > *Subject: *Re: [HCP-Users] Question about dense functional connectome > > > > We tried to open *.dconn.nii in Matlab using ft_read_cifti on an Ubuntu > system with 48 GB RAM. It used 48 GB RAM and 18 GB (out of 60 GB) swap, > then Matlab gave out of memory error! Will try to use a system with even > higher RAM. > > > > Tim: We have a list of 913 grayordinate voxels (we have their voxel > numbers/indices). These voxels are somewhat distributed in gray matter. We > are basically interested in the functional connectivity matrix just for > these 913 voxels (i.e. a 913 * 913 matrix). Is it possible to use > wb_command to select part of *.dconn.nii corresponding to the voxels of > interest? > > > > Best, > > Reza > > > > > > On Thu, Oct 25, 2018 at 3:49 AM Glasser, Matthew <glass...@wustl.edu> > wrote: > > Use swap space. > > > > Matt. > > > > *From: *<hcp-users-boun...@humanconnectome.org> on behalf of Timothy > Coalson <tsc...@mst.edu> > *Date: *Wednesday, October 24, 2018 at 4:57 PM > *To: *Reza Rajimehr <rajim...@gmail.com> > *Cc: *"hcp-users@humanconnectome.org" <hcp-users@humanconnectome.org> > *Subject: *Re: [HCP-Users] Question about dense functional connectome > > > > The cifti format doesn't support that, because it is designed to allow use > without loading the entire file into memory - there is no obvious file > organization that would allow efficient loading of a full row from a single > triangular file, when seek times are nontrivial (rotating disks). I don't > believe we have these files in any other format. I'm also not sure how > easy it would be to work on such a matrix in matlab. If wb_command is > capable of doing the operations you want, it is usually possible for it to > do them without loading the entire cifti file into memory at once. wb_view > will also display maps from it without loading the entire file into memory. > > > > DDR4 has 16GB modules available at reasonable prices, which should allow > recent computers with even 4 slots for memory to expand to 64GB. You > should also consider the "high-end desktop" platforms (threadripper, > lga2066, lga2011), which typically offer 8 memory slots, and processors > with more cores. > > > > Tim > > > > > > On Wed, Oct 24, 2018 at 3:56 PM, Reza Rajimehr <rajim...@gmail.com> wrote: > > Hi, > > > > We are trying to unzip and load dense functional connectome > (HCP_S900_820_rfMRI_MSMAll_groupPCA_d4500ROW_zcorr.zip or > HCP_S900_820_rfMRI_MSMSulc_groupPCA_d4500ROW_zcorr.zip from > > > https://www.humanconnectome.org/study/hcp-young-adult/article/announcing-release-of-s900-ptn-and-other-group-average-data) > in Matlab on a system that has 32 GB RAM. This is of course impossible due > to the fact that the files are ~33 GB in size. Due to matrix symmetry, it > would be sufficient to store only the upper or lower triangular part of a > functional connectivity matrix, reducing memory occupancy by about 50% > (Loewe et al., 2016). Does HCP provide such files that could be > memory-friendly? > > > > Best, > > Reza > > _______________________________________________ > HCP-Users mailing list > HCP-Users@humanconnectome.org > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > > _______________________________________________ > HCP-Users mailing list > HCP-Users@humanconnectome.org > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > ------------------------------ > > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you > are not the intended recipient, be advised that any unauthorized use, > disclosure, copying or the taking of any action in reliance on the contents > of this information is strictly prohibited. If you have received this email > in error, please immediately notify the sender via telephone or return mail. > > _______________________________________________ > HCP-Users mailing list > HCP-Users@humanconnectome.org > http://lists.humanconnectome.org/mailman/listinfo/hcp-users > > > ------------------------------ > > The materials in this message are private and may contain Protected > Healthcare Information or other information of a sensitive nature. If you > are not the intended recipient, be advised that any unauthorized use, > disclosure, copying or the taking of any action in reliance on the contents > of this information is strictly prohibited. If you have received this email > in error, please immediately notify the sender via telephone or return mail. > _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
