Optimized code for loop, If-statement for large dataset

I was hoping to delete some certain rows using condion. My data is in double format (790127*24) I approximate the total code need 25 hours using Run and Time which is huge. Is there any way of optimiing the script.

TIA...

n=0;
for i = 1 : length(d_A)
    
     if  any(isnan(d_A(i-n, 6))) ...
     &&  any(isnan(d_A(i-n, 7))) ...
     &&  any(isnan(d_A(i-n, 8))) ...
     &&  any(isnan(d_A(i-n, 9))) ...
     &&  any(isnan(d_A(i-n,10))) ...
     &&  any(isnan(d_A(i-n,11))) ...
     &&  any(isnan(d_A(i-n,12))) ...
     &&  any(isnan(d_A(i-n,13))) ...
     &&  any(isnan(d_A(i-n,14))) ...
     &&  any(isnan(d_A(i-n,15))) ...
     &&  any(isnan(d_A(i-n,16))) ...
     &&  any(isnan(d_A(i-n,17))) ...
     &&  any(isnan(d_A(i-n,18))) ...
     &&  any(isnan(d_A(i-n,19))) 
        d_A(i-n,:) = [];
        n=n+1;
     end
 end

Answer 1

[per isakson]

Try this

%%
ixcol = [ 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19 ];
n=0;
for i = 1 : length(d_A)
     if  all( isnan( d_A( i-n, ixcol ))) %  i-n  is a scalar
        d_A(i-n,:) = [];
        n=n+1;
     end
end

and this

%%
ixcol = [ 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19 ];
is_to_be_deleted = false( size(d_A,1), 1 );
n=0;
for i = 1 : length(d_A)
     if  all( isnan( d_A( i-n, ixcol ))) %  i-n  is a scalar
%       d_A(i-n,:) = [];
        is_to_be_deleted(i-n) = true;
        n=n+1;
     end
end
d_A( is_to_be_deleted, : ) = [];

Caveat: not tested

In response to comment:

Now it's possible to factor out the for-loop. Try this

%% Sample data
A = rand( [8,4] );
ixcol = [2,3];
A([3,5],ixcol) = nan;
A( randperm( numel(A), 9 ) ) = nan;
%%
[ A3, ix_deleted3 ] = cssm_3( A, ixcol );
[ A4, ix_deleted4 ] = cssm_4( A, ixcol );

ix_deleted3 == ix_deleted4    %#ok<NOPTS,EQEFF>

function    [ A, ix_deleted ] = cssm_3( A, ixcol )
    is_to_be_deleted = false( size(A,1), 1 );
    for jj = 1 : length(A)
         if  all( isnan( A( jj, ixcol )))  
            is_to_be_deleted(jj) = true;
         end
    end
    A( is_to_be_deleted, : ) = [];
    ix_deleted = find( is_to_be_deleted );
end
function    [ A, ix_deleted ] = cssm_4( A, ixcol )
    is_to_be_deleted = all( isnan( A( :, ixcol ) ), 2 );  
    A( is_to_be_deleted, : ) = [];
    ix_deleted = find( is_to_be_deleted );
end

it outputs

>> cssm
ans =
  2×1 logical array
   1
   1

The vectorized version, cssm_4, might not improve performance significantly, but in my opinion it makes cleaner code.

[Jhon Gray]

Wow. The second one is super fast.But there's a little bit problem here. The code would be like this.No need of i-n in this case.Thanks for helping.Take love.

ixcol = [ 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19 ];
is_to_be_deleted = false( size(d_A,1), 1 );
n=0;
for i = 1 : 1000 %length(d_A)
     if  all( isnan( d_A( i, ixcol ))) %  i-n  is a scalar
%       d_A(i-n,:) = [];
        is_to_be_deleted(i) = true;
        %n=n+1;
     end
end
d_A( is_to_be_deleted, : ) = [];

[per isakson]

I surmised that there was a problem and added the last line in bold.

It's as a bad for performance to remove one line at a time as adding one line at a time. In both cases the matrix is rewritten to memory in each operation.