isfinite() of sparse -- cleverness needed
in [Matlab]
|
From: Chad on 22 Mar 2010 13:09 Greetings, I used to have this line of code: D(~isfinite(D)) = 0; However, as the size of D has grown, I've had to go to sparse D. As a result, isfinite(D) returns an out-of-memory error. I could do something like: D=sparse(N,M); % lots-of-stuff for a=1:N*M; if ~isfinite(D(a)) D(a)=0; end end But then I get the error "FOR loop index is too large. Truncating to 2147483647. " Any clever suggestions of built in commands I'm unaware of? Thanks
From: Matt J on 22 Mar 2010 13:53 "Chad " <parishcm(a)ornl.gov> wrote in message <ho887g$kmr$1(a)fred.mathworks.com>... > Greetings, > > I used to have this line of code: > > D(~isfinite(D)) = 0; > > However, as the size of D has grown, I've had to go to sparse D. As a result, isfinite(D) returns an out-of-memory error. I could do something like: =========== You could try something like this idx=spfun(@(x) ~isfinite(x), D); D(idx)=0; Beware, however, that linear and logical indexing of large sparse matrices are vulnerable to certain bugs (as of R2009b). You might therefore want to use my RobustSparse data types instead http://www.mathworks.com/matlabcentral/fileexchange/26181-robust-sparse-data-types or else use Bruno's tools on which this is based, http://www.mathworks.com/matlabcentral/fileexchange/23488-sparse-sub-access
From: James Tursa on 22 Mar 2010 14:37 "Chad " <parishcm(a)ornl.gov> wrote in message <ho887g$kmr$1(a)fred.mathworks.com>... > Greetings, > > I used to have this line of code: > > D(~isfinite(D)) = 0; > > However, as the size of D has grown, I've had to go to sparse D. As a result, isfinite(D) returns an out-of-memory error. I could do something like: > > D=sparse(N,M); > > % lots-of-stuff > > for a=1:N*M; > if ~isfinite(D(a)) > D(a)=0; > end > end > > But then I get the error "FOR loop index is too large. Truncating to 2147483647. " > > Any clever suggestions of built in commands I'm unaware of? > > Thanks It would be fairly easy (and execute very fast) to implement this in a C mex routine. I can probably whip something up today and post it. Is your D always real or is it possibly complex? James Tursa
From: Chad on 22 Mar 2010 14:55 Matt J, Thank you for the suggestions, I will look into their possible implementations. James Tursa, Thank you for the offer! I hate to impose upon someones else's time, but if you're willing to do so, I'd be happy for your help. Real numbers, always.
From: Bruno Luong on 22 Mar 2010 15:06
"Matt J " <mattjacREMOVE(a)THISieee.spam> wrote in message <ho8aq0$3cv$1(a)fred.mathworks.com>... > "Chad " <parishcm(a)ornl.gov> wrote in message <ho887g$kmr$1(a)fred.mathworks.com>... > > Greetings, > > > > I used to have this line of code: > > > > D(~isfinite(D)) = 0; > > > > However, as the size of D has grown, I've had to go to sparse D. As a result, isfinite(D) returns an out-of-memory error. I could do something like: > =========== > > You could try something like this > > idx=spfun(@(x) ~isfinite(x), D); > D(idx)=0; > > Beware, however, that linear and logical indexing of large sparse matrices are vulnerable to certain bugs (as of R2009b). You might therefore want to use my RobustSparse data types instead > Indeed, the bug is still not fixed in 2010A: >> D=sparse(1e5,1e5,Inf) D = (100000,100000) Inf >> D(isinf(D))=0 D = (100000,100000) Inf >> D=rsparse(D) % Matt's robust sparse D= (100000,100000) Inf >> D(isinf(D))=0 % OK D= All zero sparse: 100000-by-100000 >> D=sparse(D) % convert to Matlab sparse if wanted D = All zero sparse: 100000-by-100000 % Bruno |