Vectorising nested for loops

Hi,

I’m trying to vectorise the following code snippet that works using the attached data. Unfortunately, I couldn’t succeed. Any ideas on how to deal with it?

Cheers,

Mel

k=0;
struct_data = struct('idx',[],'count',[],'size',[]);
for i = 1:size(table_data,1)
    for j = 1:list(i) 
        k=k+1;
        struct_data(k,1).idx = k;
        struct_data(k,1).count = table_data.count(i);
        struct_data(k,1).size = table_data.size(i);
    end
end

[Stephen23]

Are you intentionally storing lots of scalar cell arrays containing character vectors, or is the intent to actually just store the character vectors without the superfluous scalar cell arrays? If you only need the text, replace

table_data.size(i);

with

table_data.size{i};

https://www.mathworks.com/help/matlab/matlab_prog/access-data-in-a-cell-array.html

[Melanie VT]

Thank you for your contribution!

Answer 1

[Voss]

% unzip and load data
unzip data.zip
load data.mat

% run the loop method, for later comparison
k=0;
struct_data = struct('idx',[],'count',[],'size',[]);
for i = 1:size(table_data,1)
    for j = 1:list(i) 
        k=k+1;
        struct_data(k,1).idx = k;
        struct_data(k,1).count = table_data.count(i);
        struct_data(k,1).size = table_data.size{i}; % note this change:
                                                    % I made struct_data(k).size a char, instead of 
                                                    % a scalar cell array containing a char
    end
end
struct_data_save = struct_data;

% a "vectorized" method
idx = find(list);
struct_data = table2struct(removevars(table_data(repelem(idx,list(idx)),:),'district'));
C = num2cell(1:numel(struct_data));
[struct_data.idx] = deal(C{:});

% are the results the same?
isequal(struct_data,struct_data_save)

ans = logical
   1

[Voss]

If you really want struct_data(k).size to be a 1x1 cell array containing a character vector (as in {'venti'}), as opposed to just a character vector (as in 'venti'), you can do it with a couple of additional lines of code:

% unzip and load data
unzip data.zip
load data.mat

% run the loop method, for later comparison
k=0;
struct_data = struct('idx',[],'count',[],'size',[]);
for i = 1:size(table_data,1)
    for j = 1:list(i) 
        k=k+1;
        struct_data(k,1).idx = k;
        struct_data(k,1).count = table_data.count(i);
        struct_data(k,1).size = table_data.size(i); % scalar cell array this time
    end
end
struct_data_save = struct_data;

% a "vectorized" method
idx = find(list);
struct_data = table2struct(removevars(table_data(repelem(idx,list(idx)),:),'district'));
C = num2cell(1:numel(struct_data));
[struct_data.idx] = deal(C{:});
C = num2cell({struct_data.size});
[struct_data.size] = deal(C{:});

% are the results the same?
isequal(struct_data,struct_data_save)

ans = logical
   1

[Melanie VT]

That is awesome! Thank you so much for providing solutions for both scenarios🙏

Answer 2

[Stephen23]

S = load('data.mat')

S = struct with fields:
          list: [27×1 double]
    table_data: [27×3 table]

list = S.list

list = 27×1

0 100 60 0 0 0 0 0 0 0

tdat = S.table_data

tdat = 27×3 table
    district    count       size   
    ________    _____    __________

     {[ 1]}       91     {'venti' }
     {[ 2]}      116     {'venti' }
     {[ 3]}      112     {'venti' }
     {[ 4]}        8     {'venti' }
     {[ 5]}       23     {0×0 char}
     {[ 6]}       16     {0×0 char}
     {[ 7]}        7     {0×0 char}
     {[ 8]}        1     {0×0 char}
     {[ 9]}       28     {0×0 char}
     {[10]}       58     {'tall'  }
     {[11]}       43     {'venti' }
     {[12]}       13     {'trenta'}
     {[13]}       68     {'tall'  }
     {[14]}      108     {'grande'}
     {[15]}      102     {'venti' }
     {[16]}       76     {'venti' }

k=0;
out0 = struct('idx',[],'count',[],'size',[]);
for i = 1:size(tdat,1)
    for j = 1:list(i) 
        k=k+1;
        out0(k,1).idx = k;
        out0(k,1).count = tdat.count(i);
        out0(k,1).size  = tdat.size{i};
    end %                          ^ ^ fixed indexing
end
out0

out0 = 160×1 struct array with fields:
    idx
    count
    size

Without an explicit loop:

out1 = repelem(table2struct(removevars(tdat,'district')),list);
idx = num2cell(1:numel(out1));
[out1.idx] = idx{:}

out1 = 160×1 struct array with fields:
    count
    size
    idx

isequal(out0,out1)

ans = logical
   1

[Melanie VT]

This solution is even almost 1.5 times faster than the one above! I’m truly grateful to you all🙏🏼

This post is also an Accepted Answer with appreciation😊