Finding cell array row indices based on numeric column values

I have a large cell array keystrokes of approximate size 20000x4. Columns 1 and 3 each contain a char, while columns 2 and 4 each contain a double. For example:

>> keystrokes(378:380,:)

ans =

3×4 cell array

    {'l'      }    {[   180]}    {'e'      }    {[  69]}
    {'e'      }    {[300664]}    {'|space|'}    {[ 125]}
    {'|space|'}    {[    62]}    {'n'      }    {[2500]}

I want to find the row indices in keystrokes of occurrences of every unique combination of columns 1 and 3, where the value in column 2 is less than 100000 and the value in column 4 is less than 2000. My current code gives me the error "Undefined operator '<' for input arguments of type 'cell'.", and is shown below.

% Temporarily convert keystroke structure to a table due to unique() apparently not supporting combinations of cellarray columns.
uniqueDigraphsTable = unique(cell2table(keystrokes(:,[1 3])), 'rows');
uniqueDigraphs = table2cell(uniqueDigraphsTable);
for ii = 1:length(uniqueDigraphs)
  % Find rows containing the current unique digraph
  occurrenceIndices = find(strcmp(keystrokes(:,1), uniqueDigraphs{ii,1}) & strcmp(keystrokes(:,3), 
  uniqueDigraphs{ii,2}) & keystrokes(:,2)<100000 & keystrokes(:,4)<2000);
  ...
end

Using keystrokes{:,4}<2000 gives me this error: "Error using <. Too many input arguments." Is there a simple (and perhaps prettier) way to find the indices?

[Jan]

Prefer to post the input data such, that they can be used by copy&paste. Is keystrokes a nested cell:

kestrokes = { ...
  {'l'      }    {[   180]}    {'e'      }    {[  69]}; ...
  {'e'      }    {[300664]}    {'|space|'}    {[ 125]}; ...
  {'|space|'}    {[    62]}    {'n'      }    {[2500]}}

or a cell:

kestrokes = { ...
  'l',         180, 'e',         69; ...
  'e',      300664, '|space|',  125; ...
 '|space|',    62,  'n'        2500}

? Even typing this question need a lot of typing.

Answer 1

[Guillaume]

 find(strcmp(keystrokes(:,1), uniqueDigraphs{ii,1}) & ... split over several lines for readability
      strcmp(keystrokes(:,3), uniqueDigraphs{ii,2}) & ...
      [keystrokes{:,2}] < 100000 & ...
      [keystrokes{:,4}] < 2000)

or

 find(strcmp(keystrokes(:,1), uniqueDigraphs{ii,1}) & ... split over several lines for readability
      strcmp(keystrokes(:,3), uniqueDigraphs{ii,2}) & ...
      cell2mat(keystrokes(:,2)) < 100000 & ...
      cell2mat(keystrokes(:,4)) < 2000)

In essence you have to transform your cell columns into numeric matrices.

[Piddy]

Thanks a lot! Your cell2mat solution gives the results I'm looking for. The first solution seems to have sort of looping problem though. It produces a very large vector where the first elements are the correct indices, but following those are indices that exceed the length of the keystrokes array.

For example, when keystrokes is a 24894x4 cell, part of its output for a specific row in uniqueDigraph looks like this:

K>> length(occurrenceIndices) 
  ans =

      158473

K>> occurrenceIndices(1:15)

ans =

         591
         677
        1090
        2247
        2578
        2912
        3227
       25485
       25571
       25984
       27141
       27472
       27806
       28121
       50379

The first 7 values are correct, but the rest are too large. 24894 + 591 = 25485 though, and 24894 + 677 = 25571 etc.

Answer 2

[Jan]

The cell is not useful for these comparisons. Converting is to a table is the next indirection. Easier:

% Store strings in one cell string:
Strings  = keystrokes(:, [1, 3]);
uStrings = unique(Strings, 'rows'); 

% Store numbers in a numerical array:
Values = cell2mat(keystrokes(:, [2, 4]));

% Move the check of the values out of the loop for performance:
match = (Values(:, 1) < 100000 & Values(:, 2) < 2000);
for ii = 1:length(uStrings)
  occurrenceIndices = find(strcmp(Strings(:,1), uStrings{ii, 1}) & ...
                           strcmp(Strings(:,2), uStrings{ii, 2}) & ...
                           match);
  ...
end

This would be faster, if you use the 2nd and 3rd output of unique() also:

[uStrings, iString, iUniq] = unique(Strings, 'rows');
match = (Values(:, 1) < 100000 & Values(:, 2) < 2000);
for ii = 1:length(uStrings)
  occurrenceIndices = find(iUniq == ii & match);
  ...
end

[Piddy]

Thank you! There is still an issue though. The following line produces this warning: "The 'rows' input is not supported for cell array inputs."

[uStrings, iString, iUniq] = unique(Strings, 'rows');

Does this tie into your comment asking whether or not keystrokes is a nested cell? I didn't produce the keystrokes variable myself, but I'm fairly sure that it is not nested. I checked using class():

class(keystrokes{1,1})
ans = 'char'

I also think that if it were nested, the example command I showed in my original question would have produced an output like this:

>> keystrokes(378:380,:)

ans =

3×4 cell array

    {1×1 cell}    {1×1 cell}    {1×1 cell}    {1×1 cell}
    {1×1 cell}    {1×1 cell}    {1×1 cell}    {1×1 cell}
    {1×1 cell}    {1×1 cell}    {1×1 cell}    {1×1 cell}

I could of course be mistaken.

[Guillaume]

Annoyingly, unique (and ismember) do not support the 'row' option with cell arrays even if it is a cell array of char arrays. If you have matlab R2016b or later, you can convert the cell array of char arrays into a string array which can be used with unique and the 'row' option:

unique(string(keystrokes(:, [1 3])), 'rows')