Explode cell that are into another cell

I have a cellarray data like in the picture : each cell contains cellarray of strings

I am opening all all the cells with this way

counter=0;
for ind=1:length(data)
  tmp=cell2str(data{ind,1});
  for k=1:size(tmp,1)
    counter=counter+1;
    tmp2=textscan(tmp(k,:),'%s%s%s%s%s%[^\n\r]','Delimiter', ' ');
    for j=1:6
      if isempty(tmp2{j})==0
        Raw(counter,j)=tmp2{j};
      end
    end
    clear tmp2 j
  end
  clear k tmp
end

The results is correct but is there a better/faster way to do it ?

Using parfor, or other technics

Thank you in advance

[dpb]

Attach small sample dataset and what do you want the result to be?

[NirE]

I am joining a small piece of data, 10 rows but in fact it has 38956 rows.

The length of the cells that are inside the main cell can differ. I hope that I am clear enough

You can just run the script that I wrote it is working but very slowly

Is there a way tu use parfor or other thing in order to accelerate the process?

[Luna]

cell2str does not work for me, which version you are using?

[NirE]

Matlab R2017b

[dpb]

Must be in some TB, then, it's not in base R2017b...

Again, what's the desired output? That doesn't seem to make sense reading the code; your format statement has 5 strings but some "records" have many more fields than that...

Well, that's not it either, maybe FEX submittal? A search of online help doesn't find it, either.

Not knowing what, precisely, the cell2str function actually returns it's hard to guess exactly what the result that "works" really is without more effort than have time to spend...help us help you.

[Luna]

What size should be your output? Are you expecting 64x1 cell array?

[NirE]

the str2cell function just return a string vector with the number of line that we had in the cell.

for the format it is exactly what i want 6 parts with different length.

i hope that i am helping

[Jan]

Note: Omit the useless clear commands. They will waste time only here.

[dpb]

Well, no...helping would be to show us what you really, really want instead of just describing it that we can't reproduce.

Where did you find the function? SHOW us!!!

Answer 1

[Jan]

Start with a pre-allocation:

Len = cellfun('prodofsize', data);
Raw = cell(sum(Len), 6);
c   = 0;
for ind = 1:numel(data)
  tmp = data{ind};
  for k = 1:numel(tmp)
    c    = c + 1;
    tmp2 = strsplit(tmp{k}, ' ');
    for j = 1:numel(tmp2)
      Raw{c, j} = tmp2{j};
    end
  end
end

I cannot open your MAT file currently, so I guess, what it might contain. I guessed also, that cell2str can be avoided by scanning the cell element directly. I assume that Raw should be a cell array. All these assumptions can be wrong. If you post a small input as code and the wanted output, less guessing is required.

[NirE]

Your code is reducing mine 3 times thanks a lot.

Can you just explain what is doing 'prodofsize' ?

[Jan]

@Nir Eliezer: 'prodofsize' is explained in the docuemtation: doc cellfun. It is equivalent to:

Len = cellfun(@numel, data);

but much faster. If you provide a function handle to cellfun, it calls the Matlab engine for each element of the cell, while using the strings like 'length', 'prodofsize' and 'isclass' accesses the cell elements directly inside the cellfun core. Although the speed of cellfun might be negligible in your case, it is a good programming practize to use the most efficient code.

By the way, 'numel' would be much nicer than 'prodofsize'.

Maybe this is slightly faster:

Len = cellfun('prodofsize', data);
Raw = cell(sum(Len), 6);
c   = 0;
for ind = 1:numel(data)
  tmp = data{ind};
  for k = 1:Len(ind)
    c    = c + 1;
    tmp2 = strsplit(tmp{k}, ' ');
    Raw(c, 1:numel(tmp2)) = tmp2;
  end
end

[NirE]

Jan one more question is there a way to parallelize your piece of code that I could use parfor ?

[Jan]

Yes, a parallelizaion should be very straigh forward. Did you try it?

[NirE]

Will try and tell you how it increase or not

Answer 2

[dpb]

OK, I overlooked the regular expression in the format string that sucks up all of those extra blanks at the end of the odd-man-out records...

To dereference the cell content in each cell requires two levels snce textscan isn't cell-string aware. split doesn't cut it here because there's not a unique delimiter that defines the fields desired; hence the above...

You can try the following and see if the lack of preallocation shows up as a performance hit with the size; oftentimes it'll fool you and not be too bad...

fnTS=@(s) textscan(s,'%s%s%s%s%s%[^\n\r]','Delimiter', ' ');
res=[];
for i=1:length(data)
    res=[res;cellfun(fnTS,data{i},'uni',0)];
end
res=cat(1,res{:});

The above yields a 64x6 cell array...

I'd have to think of the bestest way to be able to build the array directly w/o the intermediary second cell array to not be dynamically catenating the output.

ADDENDUM:

res(cellfun(@isempty,res))={''};
>> string(res)
ans = 
  11×6 string array
    "1"    "EventDataLogNewFile"    "DataEventTime"             "TypeSecondsSinceEpoch"    "1546725641"                 ""                    
    "1"    "EventDataLogNewFile"    "DataEventTime"             "TypeFormattedDate"        "Sun"                        "Jan  6 00:00:41 2019"
    "1"    "EventDataLogNewFile"    "DataReportingSubsystem"    "TypeString"               "datalogger"                 ""                    
    "1"    "EventDataLogNewFile"    "DataInstrumentID"          "TypeString"               "00:01:05:19:CF:30"          ""                    
    "1"    "EventDataLogNewFile"    "DataEntityName"            "TypeString"               "mc16"                       ""                    
    "4"    "EventREAD"              "DataEventTime"             "TypeSecondsSinceEpoch"    "1546725657"                 ""                    
    "4"    "EventREAD"              "DataEventTime"             "TypeFormattedDate"        "Sun"                        "Jan  6 00:00:57 2019"
    "4"    "EventREAD"              "DataReportingSubsystem"    "TypeString"               "pc"                         ""                    
    "4"    "EventREAD"              "DataEntityID"              "TypeString"               "Dev_CLPC_PressureGauge1"    ""                    
    "4"    "EventREAD"              "DataReading"               "TypeUnitLessNumber"       "34729"                      ""                    
    "4"    "EventREAD"              "DataEventDuration"         "TypeSec"                  "0.000686859"                ""                    
>> 

for just doing the first two elements in the for...end loop instead of all for brevity.

Answer 3

[Luna]

I was assuming the same 64x9 cell. Here is my solution gives the same result with Jan's:

cellArray = cellfun(@(x) strsplit(x(:,:),' '), vertcat(data{:}), 'UniformOutput',false);
for i = 1:numel(cellArray)
    for j = 1:numel(cellArray{i})        
        raw{i,j} = cellArray{i}{j} ;       
    end    
end