Generate new variables based on expressions in structure array using loops -- example provided need help to fix

Can someone help direct me on how to get this to work? For example, I would like to create a new variable in the 'var' structure array that identifies the last year for each cell array and then assigns the last value in the data variable.

var.time = cell(4,1);
var.time{1,1}={'2011';'2012';'2013';'2014'};
var.time{2,1}={'2011';'2012';'2013'};
var.time{3,1}={'2011';'2012'};
var.time{4,1}={'2011';'2012';'2013';'2014';'2015'};
var.data{1,1} = {1;2;3;4};
var.data{2,1} = {1;2;3};
var.data{3,1} = {1;2};
var.data{4,1} = {1;2;3;4;5};

[m,n] = size(var.time);

for j = 1:m;
varName{j,1} = ['var.',var.time{j,1}{end,1},'{',num2str(j),'}'];
var.(varName{j,1}) = var.data{j,1}{end,1};
end

Answer 1

[Guillaume]

I'm a bit confused by what you're asking as you seem to have the answer in your question. It's just that the name you generate is not a valid field name because of the dot.

To create / use a field whose name you generate from an expression, use dynamic field names which are enclosed in brackets:

for year = [2011 2012 2013]
   fname = sprintf('Min%d', year);
   s.(fname) = year;
end

If you end up creating field names that are not valid, you can transform them into valid field names with genvarname.

[Jeff]

Hi Guillaume / Geoff:

Here is another bit of code that I hope helps illustrate better what i am trying to achieve. My goal is to automatically create a variable into the structured array that is based on an expression (so the year in the field name changes based on unique year indexed)

=========================================

    var.time = cell(4,1);
    var.data = cell(4,1);

    var.time{1,1} = (now:-1:(now - ((365*3)+1)))';
    var.time{2,1} = (now:-1:(now - ((365*2)+1)))';
    var.time{3,1} = (now:-1:(now - ((365*5)+1)))';
    var.time{4,1} = (now:-1:(now - ((365*4)+1)))';

    [m,n] = size(var.time);

 for j = 1:m
    dates_1{j,1}= datestr(var.time{j,1},'yyyy-mm-dd');
    dates{j,1} = datevec(dates_1{j,1});
    var.year{j,1} = dates{j,1}(:,1);
    var.data{j,1} = rand(length(var.year{j,1}),1,'double');
    var.uniqueYears{j,1} = unique(var.year{j,1});
    var.min2014{j,1} = min(var.data{j,1}([var.year{j,1} == 
    var.uniqueYears{j,1}(end,1)]));
 end

clear dates_1 dates j m n;

% I would like to have a loop that automatically creates a variable for % each year and to assign the minimum value for that year. Above i % manually had to name min2014 but I would like this to be related to the % uniqueYears i am indexing and to name it automatically. Is this % possible?

==========================================

[Guillaume]

Please format your code properly, as otherwise it's very difficult to understand.

Isn't the answer I gave what you want? That is, use dynamic field names. To use your example, replace your var.min2014... line with:

for year = unique(var.year{j})
   s.(sprintf('min%d', year)){j, 1} = min(...)
end

[Jeff]

Hi Guillaume,

sprintf looks like that might be what I was missing. Let me give that a try and follow back up. Sorry I only realized now how to format the code. Hope that looks better.

Thanks again, Jeff

[Jeff]

Hi Guillaume,

this is definitely working however i am hitting another new error and have been trying to work around it but can't seem to hit it yet. Since my goal is to automate the fieldnames I need to have a condition that creates an NaN when the index is beyond the length of the cell array. If you run below you will see that an error occurs on j=2 because I am indexing beyond the length of the cell array {3,1}. Can you help to show me how I can insert a NaN when the index is outside the length of the uniqueYear variable.

var.time = cell(4,1);
var.data = cell(4,1);

var.time{1,1} = (now:-1:(now - ((365*3)+1)))';
var.time{2,1} = (now:-1:(now - ((365*2)+1)))';
var.time{3,1} = (now:-1:(now - ((365*5)+1)))';
var.time{4,1} = (now:-1:(now - ((365*4)+1)))';

[m,n] = size(var.time);

for j = 1:m
    dates_1{j,1}= datestr(var.time{j,1},'yyyy-mm-dd');
    dates{j,1} = datevec(dates_1{j,1});
    var.year{j,1} = dates{j,1}(:,1);
    var.data{j,1} = rand(length(var.year{j,1}),1,'double');
    var.uniqueYears{j,1} = unique(var.year{j,1});

      var.(sprintf('min%d', var.uniqueYears{j,1}(end,1))){j,1}   = min(var.data{j,1}([var.year{j,1} == var.uniqueYears{j,1}(end,1)]));
      var.(sprintf('min%d', var.uniqueYears{j,1}(end-1,1))){j,1} = min(var.data{j,1}([var.year{j,1} == var.uniqueYears{j,1}(end-1,1)]));
      var.(sprintf('min%d', var.uniqueYears{j,1}(end-2,1))){j,1} = min(var.data{j,1}([var.year{j,1} == var.uniqueYears{j,1}(end-2,1)]));
      var.(sprintf('min%d', var.uniqueYears{j,1}(end-3,1))){j,1} = min(var.data{j,1}([var.year{j,1} == var.uniqueYears{j,1}(end-3,1)]));

end

clear dates_1 dates j m n;

[Guillaume]

A few things first,

First of all, you should have accepted the answer and started a new question.

Secondly, you're using vectors (not matrices) so most indexing can just be done with varname{index} instead of varname{index, 1}. I find it much clearer. The only difference is when you create varname with _varname{index}, it'll create a row vector instead of a column vector.

Thirdly, it's not clear to me what are temporary variables and what are actual variables you want to store. Is var.year important? What about var.uniqueYears?

Fourth, dates_1 and dates are clearly temporary. Is there any reason to make them a cell array? You can also dispense with dates_1 altogether and calculate dates directly from var.year.

With all that said, I think this is what you want:

var.time = cell(4, 1);
var.data = cell(4, 1);

var.time{1} = (now:-1:(now - ((365*3)+1)))';
var.time{2} = (now:-1:(now - ((365*2)+1)))';
var.time{3} = (now:-1:(now - ((365*5)+1)))';
var.time{4} = (now:-1:(now - ((365*4)+1)))';

var.year = cell(size(var.time));
for j = 1:numel(var.time)
  date = datevec(var.time{j}); %no need to go through datestr.
  var.year{j} = date(:, 1); 
  var.data{j} = rand(numel(var.year{j}), 1); %double by default
end

for year = unique(vertcat(var.year{:}))' %vertcat since you're using column vectors.
   var.(sprintf('min%d', year)) = cellfun(@(y, d) min(d(y == year)), var.year, var.data, 'UniformOutput', false)';
   %note that the line above will put empty matrices for missing years
   %if you really do want NaNs uncomment the following line
   %var.(sprintf('min%d', year)){cellfun(@isempty, var.(sprintf('min%d', year)))} = NaN;
end

clear dates j year

[Jeff]

Thanks Guillaume for all your help.

For the question since it's two parts allow me to still create a second question, if you would like (sorry wasn't thinking).

Thank-you for your point about indexing vector arrays, i guess it's a bad habit but still good to know.

Yes to the temporary variables, actually the whole script is a dummy script based on another file I am working with. Knowing the unique years for each cell array would be something needed, probably don't need to actually store it but would need to know what years are there to assign in the fieldname -- but seeing your answer i guess clearly not :)

Thanks for explaining/showing the alternative date method!!

Overall, thank-you for teaching me some more Matlab!! Program works perfectly ... now to transfer and apply what you taught me.

Cheers

Answer 2

[Geoff Hayes]

Jeff - see the generate field names from variables for details on creating new fields for a structure given a string variable.

In your case, you could do something like

 for k=1:m
     % create the new field name
     fieldName = ['y' var.time{k,1}{end}];

     % add the field to the structure with the correct value
     data.(fieldName) = var.data{k,1}{end};
 end

After the code has completed, var looks like

 var = 
     time: {4x1 cell}
     data: {4x1 cell}
    y2014: 4
    y2013: 3
    y2012: 2
    y2015: 5

Note that the new field names need to be prefixed with a non-numeric (and non-symbolic) character like 'y'.

The above shows that you can do this, but is this something that you really want? What is the intent behind showing the data in this way?

An alternative would be to create a 4x2 cell array where the first column represents the year and the second column the last value (or a 4x2 matrix since the time and data elements are numeric)

 var.lastYearData = cell(m,2);

 for k=1:m
     % assign the time
     var.lastYearData{k,1} = var.time{k,1}{end};

     % assign the data
     var.lastYearData{k,2} = var.data{k,1}{end};
 end

Now, after the code has completed, var looks like

 var = 
            time: {4x1 cell}
            data: {4x1 cell}
    lastYearData: {4x2 cell}

with

 var.lastYearData

 ans = 
    '2014'    [4]
    '2013'    [3]
    '2012'    [2]
    '2015'    [5]

I think that this alternative is cleaner that the first suggestion.

Try either option and see what happens!

[Jeff]

Hi Geoff, thanks for your help although this is not exactly what I am trying to do in my own dataset. I think my example and question might have been a bit unclear.

Basically I have structured array with hundreds of cell arrays of different lengths. I tried to demonstrate that with the test data provided:

var = {4x1} where each cell has different lengths (for example {1x1} of the structure array contains a {4x1} cell array,{2x1}{3x1} etc.)

what I am doing is taking say the minimum value of each cell array within the structure, creating a scalar for each cell array.

the variable (fieldname) containing that scalar i am trying to automate base on an expression. The minimum value is extracted by year so I am trying to automate the fieldname

The way I currently do the same thing is by manually naming the new field name and then assigning my expression for across all cell arrays. For example:

for j=1:m MyDate.Min2013{j} = expression end

the above adds Min2013 to each of my cell arrays containing a scalar value.

What I would like to automate is the fieldname, which I am currently naming manually as say 'Min2013', but where the year in the fieldname is now determined based on an expression.

Is this possible to do?

-- I will try to get a better example if this is not clear

[Geoff Hayes]

Yes, a better example is needed especially as you switched from maximum to minimum, and have mentioned an expression.

what I am doing is taking say the minimum value of each cell array within the structure, creating a scalar for each cell array.

Okay, so this would be (from your above example) 2011 repeated four times: [2011 2011 2011 2011]. How do you expect to handle the case, like this one, where you have four identical scalars? Are you assuming that each cell array starts from the same year, so all will have the same minimum year?

 minYear = Inf;
 for k=1:m
     val = min(str2num(char(var.time{m,:})));
     if val<minYear
         minYear = val;
     end
 end

the variable (fieldname) containing that scalar i am trying to automate base on an expression. The minimum value is extracted by year so I am trying to automate the fieldname

What is the expression? Or is this the data part of the var?

The way I currently do the same thing is by manually naming the new field name and then assigning my expression for across all cell arrays. For example:

Something like

 fieldName = ['min' num2str(minYear)];
 var.(fieldName) = cell(m,1);
 for k=1:m
     var.(fieldName){k,1} = var.data{k}{1};
 end

What I would like to automate is the fieldname, which I am currently naming manually as say 'Min2013', but where the year in the fieldname is now determined based on an expression.

So the above code assumes that the expression is minimum, is that correct? Now you want to use some other expression instead?