So I wanted to split the 1X15 cell to different cells.

Yes this is a screenshot of a screenshot.

So I have a 1x15 cell

Each (cell)column is one channel so a total of 15 channels.

2nd screenshot is a 1st channel with 2 states.

So now i wanted to put each state of one channel into a seperate matrix or a cell(preferably a cell).The state changes from rest to active every 200th iteration and vice versa.

So new cell of channel 1 would contain 1 cell for with values for rest block and next cell for active block and so on.

so repeat this for 15 channels.

I have attached mat file of my data.

[Guillaume]

What do you want to do with that data?

Personally, I'd go the other way, and put everything into one table with 3 columns: Channel, hBo, state

Most likely, whatever it is you want to do will be easier with just one table than cell array of cell arrays of cell arrays.

[CalebJones]

Guillaume

I wanted to split it because i wanted calculate mean,slope,kurtosis,peak which i will use as my feature set for training my model.

[CalebJones]

Guillaume any thoughts?

Answer 1

[Guillaume]

If you want to perform group statistics, the simplest is indeed to put everything in just one table:

allresults = vertcat(result{:});  %concatenate everything into one table
allresults.channel = repelem((1:numel(result))', cellfun(@height, result));  %and add a channel column

It's then trivial to calculate group statistics with groupsummary or grpstats:

stats = groupsummary(allresults, {'state', 'channel'}, {'mean', @kurtosis});  %not sure what you mean by slope and peak

[CalebJones]

Guillaume

Doesn't do what i expected it to do.

Thanks anyways.

appreciate if you could be me out further.

[Guillaume]

appreciate if you could be me out further

Clearly, I haven't understood what you want, so I need more details in order to help further.

[CalebJones]

If you have a look at this image. Like wise even i have extracted HbO which is oxy hemoglobin and HbR which is deoxy Hemoglobin.

And in my dataset is a task like rest active above image shows rest and tap.

Above image plots only 1 channel signal , likewise there at 15 channels.

My dataset(attached above) has calculated HbO Values and labeled things as rest and active for each channel.

So now i wanted to train my model such that it predicts rest and active states with decent accuracy using features like slope,mean,kurtosis,skewness,etc.

https://sci-hub.tw/https://www.ncbi.nlm.nih.gov/pubmed/27252637

https://sci-hub.tw/https://ieeexplore.ieee.org/document/6181455

Guillaume hope now you get a gist of whats happening?

Thank you

[Guillaume]

If it's stat per each rest and active period you want, then you simply need to add another variable to identify each of these period:

%this is as before:
allresults = vertcat(result{:});  %concatenate everything into one table
allresults.channel = repelem((1:numel(result))', cellfun(@height, result));  %and add a channel column
%now add a variable whose value is identical for each continuous rest/active period:
allresults.eventid = cumsum([1; diff(findgroups(allresults.state)) ~= 0]);

You can now use that variable for grouping:

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @kurtosis});  %not sure what you mean by slope and peak

You can specify your own statistical function to pass to groupsummary. It will be called once for each distinct group, with the HbO vector of values for each group.

[CalebJones]

Thank You Guillaume

It will be called once for each distinct group, with the HbO vector of values for each group.

Could you elaborate with an example?

Is it like @kurtosis_allresult would call---> kurtosis_allresult = kurtosis(allresult,0,'all');

Similarly...

Skweness_allresult = skewness(allresult);

slope_allresult = diff(allresult);

[Guillaume]

No, I've given you an example above:

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @kurtosis});  %not sure what you mean by slope and peak

Since I don't have the stats toolbox (with kurtosis), I can't show you the result for that, but here is an example with a made-up function:

function s = slope(values)
    %fit a linear polynomial to the data, and return the slope of the line
    if ~isempty(values)
        p = polyfit(1:numel(values), values', 1);
        s = p(1);
    else
        s = NaN;
    end
end

and using it on your data:

%merge all into one table with extra columns:
allresults = vertcat(result{:});  %concatenate everything into one table
allresults.channel = repelem((1:numel(result))', cellfun(@height, result));  %and add a channel column
allresults.eventid = cumsum([1; diff(findgroups(allresults.state)) ~= 0]);

%get stats: mean and slope
stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @slope});  %not sure what you mean by slope and peak
stats.Properties.VariableNames{end} = 'slope_HbO'; %better variable name that what matlab autogenerate

head(stats) returns:

>> head(stats)  %first 8 rows
ans =
  8×6 table
    state     channel    eventid    GroupCount        mean_HbO              slope_HbO     
    ______    _______    _______    __________    _________________    ___________________
    active       1          2          200        -7.67231127895478     -0.091165802772305
    active       1          4          200         35.5962270614642      0.076534807760347
    active       1          6          197        -19.7591859046008      0.252152452160729
    active       2          8          200        -15.9692390786319      0.235640586381289
    active       2         10          200         20.5796697869384    -0.0664683822345007
    active       2         12          197        -17.5921264661321      0.200172452731579
    active       3         14          200         7.17042032876279    -0.0507776796191182
    active       3         16          200          12.637946834818     -0.218112096344478

[CalebJones]

Guillaume Thanks a lot....

But one question why are channels repeating? number of Rest and Active per channel add up to the correct value which is 1197 which is correct.

But could you explain why ?

[CalebJones]

Guillaume ,

Okay say i used only mean and slope for training i should be excluding eventid and channel right?

Obvioulsy state should be my response (rest or active).

And run the classifier with Groupcount, mean and slope right?

[Guillaume]

why are channels repeating

You get one row per unique combination of channel, state and eventid.

With regards to the classification, I'm afraid I can't help with that, I don't have the toolbox. If you don't care which channel the data come from, I assume you should remove it from the classification. You obviously don't care about the eventid anymore, its only purpose was to allow grouping into consecutive identical states.

[CalebJones]

Guillaume

I analysed it step by step.Now it makes sense.Your a genuis mate.

I was looping to do all this and it was driving me nuts.Your code was just consise and to the point.

[CalebJones]

Guillaume,

I was calculating variance and its throwing this error----

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @slope_HbO,@variance_HbO});

the function variance_HbO contains this----

function  vari = variance_HbO(values)
    %fit a linear polynomial to the data, and return the slope of the line
    if ~isempty(values)
        v = var(1:numel(values), values', 1);
        vari = v(1);
    else
        vari = NaN;
    end

error-------

Error using groupsummary (line 522)
Unable to apply method 'fun2' to data variable HbO.

Caused by:
    Error using var (line 192)
    W must be a vector of nonnegative weights, or a scalar 0 or 1.

[Guillaume]

You don't need a separate function for that. var is supported by groupsummary:

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @slope_HbO, 'var'});

Unless you do want a weight of 1. In which case if you use a separate function it's as simple as:

function vari = variance_hBO(values)
    vari = var(values, 1);  
end

You could also use an anonymous function

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @slope_HbO, @(v) var(v, 1)});

[CalebJones]

Guillaume,

Thanks that worked...

I was trying to include peak as well.

stats = groupsummary(allresults, {'state', 'channel', 'eventid'}, {'mean', @slope_HbO, @(v) var(v, 1), @(p) findpeaks(p)});

but it was throwing an error,

Error using groupsummary (line 522)
Unable to apply method 'fun3' to data variable HbO.

Caused by:
    Error using findpeaks
    Expected Y to be nonempty.

If I try k = findpeaks(allresults.HbO); then size of k is 229x1 so will that be an issue while grouping?Is there a way around this ?

[CalebJones]

Guillaume,

Similarly could you help me out with skewness and kurtosis as well.

[Guillaume]

You have two issues there.

As an implementation detail, groupsummary first call the function with an empty vector (not sure why) and findpeak doesn't like that. That's the error you're getting.

The second issue is that groupsummary wants scalar output out of the function and that's not going to be the case with findpeaks.

You could fix both problems with this function (untested):

function out = my_findpeaks(values)
    if isempty(values)  %function may be called with empty vector by groupsummary, which is a problem for findpeaks
        out = {};  %return scalar cell with nothing in it
    else
        out = {findpeaks(values)};  %wrap output of findpeaks in a scalar cell
    end
end

But do you really want to return all the peaks? it may be better to extract just one or two peaks from the data in the above function.

[CalebJones]

Guillaume,

function out = my_findpeaks(values)

if isempty(values) %function may be called with empty vector by groupsummary, which is a problem for findpeaks

out = {}; %return scalar cell with nothing in it

else

out = {findpeaks(values)}; %wrap output of findpeaks in a scalar cell

end

end

Unfortunatly, this doesn't work.

And yes perhaps one or two peaks would be better than all.

[Guillaume]

Unfortunatly, this doesn't work.

Because ?

[CalebJones]

Guillaume,

Error using groupsummary (line 522)

Unable to apply method 'fun3' to data variable HbO.

Caused by:

Error using findpeaks>parse_inputs (line 206)

Data set must contain at least 3 samples.

[Guillaume]

if numel(values) < 3

instead of

if isempty(values)

should fix it.

[CalebJones]

Ah great it worked....Thank You so much Guillaume