sort X and Y columns according to a repeated string in a 3rd column and scatter plot

Hi, I have 3 columns (in a struct array called "data") columns are; "subject", "timepoint", "perimeter" . How can can I use something like gscatter to group the "person" column (there are 8 people) - I can get it to work for one person but am having trouble making it work for all 8 people . Thanks

 x =[]; y = []; 

for i = 1:length(data)
    if strcmp(data(i).subject,'Person1')

         x(i) =  data(i).timepoint; 
         y(i) =  length(data(i).perimeter); 

      end

end
scatter(x, y);

[cgenes]

I guess people are misunderstanding me - I want to sort column 2 and 3 (time and perimeter) by a third column "subject" where each row has person1, person1, person1, person1 . . . . person 2, person 2, person 2 . . . etc etc - - sorry to be confusing!!

[Guillaume]

Do you actually need to sort the structure or to plot a scatter plot by group?

[cgenes]

OMG - just found the unique() function - - boom!

[Stephen23]

http://xyproblem.info/

http://meta.stackexchange.com/questions/66377/what-is-the-xy-problem

Answer 1

[Guillaume]

This should work (see Access elements of a nonscalar struct array)

gscatter([data.timepoint], [data.perimeter], {data.subject})

But, I would recommend you convert you struct array into a table which is easier to work with:

tdata = struct2table(data);
gscatter(tdata.timepoint, tdata.perimeter, tdata.subject)

edit: So it seems you just want to sort the data and actually don't care about a scatter plot. Again, this is much easier if you convert to a table:

tdata = struct2table(data);
tdata = sortrows(data, 'subject');

With a structure:

 [~, order] = sort({data.subject});
 sorteddata = data(order);

[cgenes]

I've found this which is sort of my problem https://uk.mathworks.com/matlabcentral/answers/55167-how-to-plot-x-y-coordinates-which-correspond-to-elements-which-are-repeated-at-least-once

although I don't quite get the answer yet - - thanks for you comments anyway!!

[Guillaume]

Doesn't gscatter do what you want?

Otherwise, is it something like this?

tdata = struct2table(data);
[group, name] = findgroups(tdata.subject);
figure; hold on;
splitapply(@plot, tdata(:, {'timepoint', 'perimeter'}), group);

[cgenes]

Yes this works! thanks very much - - also . .

gscatter([tdata.timepoint],[tdata.perimeter], {tdata.subject})

this works just fine too - - I think I was trying to use the same thing on the structured array without turning into a table first and it didn't like it

just one more thing - - I want to further split my groups into "young" and "old" people, and have 2 subplots - - I can do it with contains() as the person name has either "Y" or "E" I've tried this but it doesn't work - note i'm doing the contains() on the structured array not the table as I couldn't get it to work on the table

for i = 1:length(data)
    if contains({data(i).subject}, 'Y') == 1
        data(i).Ay = data(i).subject; 
    elseif contains(data(i).subject, 'E') == 1
        data(i).Ae = data(i).subject; 
    end 
end

%%

tdata = struct2table(data);
[group, ID1, ID2] = findgroups(tdata.Ay, tdata.Ae);
figure; hold on;
splitapply(@scatter, tdata(:, {'timepoint', 'perimeter'}), group);
legend(ID1, ID2)

[Guillaume]

There are many ways to do what you want. I would actually create a new column for the age:

tdata.IsYoung = contains(tdata.subject, 'Y');

Or to make it more fancy, you could convert the logical into categorical:

tdata.Age = categorical(contains(tdata.subject, 'Y'), [true false], {'Young', 'Old'});  %instead

For the plotting, you'd use the original findgroups and filter the whole table for each subplot:

[group, name] = findgroups(tdata.subject);
figure; hold;
subplot(1, 2, 1);
splitapply(@scatter, tdata(tdata.IsYoung, {'timepoint', 'perimeter'}), group(tdata.IsYoung));
subplot(1, 2, 2);
splitapply(@scatter, tdata(~tdata.IsYoung, {'timepoint', 'perimeter'}), group(~tdata.IsYoung));

If using the categorical column, replace tdata.IsYoung by tdata.Age == 'Young'.

[cgenes]

Hi thanks very much for this - - the new columns are made if i try plotting the logical way it says ....

Error using splitapply (line 108) For N groups, every integer between 1 and N must occur at least once in the vector of group numbers.

and the logical way it says . . .

Error using splitapply (line 96) The data variables must have the same number of rows as the vector of group numbers. The group number vector has 37 row(s), and data variable 1 has 19 row(s).

thanks for all your help on this by the way!! I think i'll just create two new tables one for young and one for old

[Guillaume]

"For N groups, every integer between 1 and N must occur at least once in the vector of group numbers"

That's a limitation imposed by splitapply that I wasn't aware of. So, yes my method is not going to work.

Splitting the table in two

isyoung = contains(tdata.subject, 'Y');
tyoung = tdata(isyoung, :);
told = tdata(~isyoung, :);

and doing findgroup and splitapply on each is probably the simplest indeed.

Answer 2

[Massimo Zanetti]

Get the "subject" column in one array containing data of all people, as follows:

allsubjects = data(:).subject

[cgenes]

not quite sure what you mean - this just is a character array containing "person1"

[Guillaume]

data(:).subject (or simply data.subject) returns a comma separated list. With the answer given, only the first element of the list is assigned (to allsubjects), so the effect of the above is actually:

allsubjects = data(1).subject;

The correct syntax should have been:

allsubjects = {data.subject};

to convert the list into a cell array of strings. This is what I have done in the third argument to gscatter in my answer (and what Massimo is trying to say, I assume).

To avoid dealing with comma-separated lists, conversion to cell arrays or vectors (depending on type), etc. convert your structure into a table as I've also shown.

[Massimo Zanetti]

Sorry , I forgot the { }