How to randomly select the datapoints in a vector based on percentage for each group?

Divide the data values into groups using the percent distributions of the data values: Group 1=25%, Group 2=30%, Group 3=20%, Group 4=25%.

Answer 1

[DGM]

There has to be a smarter way than this, but I guess this is one idea.

% some fake data
x = randn(1000,1);

% the percentiles (should be a unit sum)
prct = cumsum([0.25 0.30 0.20 0.25])

prct = 1×4

0.2500 0.5500 0.7500 1.0000

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

% where they lie in data units
pval = prctile(x,prct*100)

pval = 1×4

-0.7367 0.1090 0.6688 3.1929

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

% bin the data
nbins = numel(prct);
xbinned = cell(nbins,1);
for k = 1:nbins
    switch k
        case 1
            mask = x <= pval(k);
        case nbins
            mask = x > pval(k-1);
        otherwise
            mask = x > pval(k-1) & x <= pval(k);
    end
    xbinned{k} = x(mask);
end
xbinned

xbinned = 4x1 cell array
    {250x1 double}
    {300x1 double}
    {200x1 double}
    {250x1 double}

... that's assuming I understand the question correctly.

[DGM]

... I just realized the question said "randomly", so I probably completely misinterpreted the question.

Answer 2

[Star Strider]

I’m not certain iif you want to apportion them as they exist in the original vector, or if you want to apportion them by ascending value (essentially their percentile ranks).

Here are two methods of apportioning them —

x = randn(153,1);
L = numel(x);
gv = [25 30 20 25];
g1 = round(gv*L/100);
xg1 = mat2cell(x, g1, size(x,2))                                %  Option #1: Apportion Without Sorting

xg1 = 4x1 cell array
    {38x1 double}
    {46x1 double}
    {31x1 double}
    {38x1 double}

[xs, sidx] = sort(x);                                           % Sort Ascending
xg2idx = mat2cell(sidx,  g1, size(x,2))                         % Collect Sort Indices

xg2idx = 4x1 cell array
    {38x1 double}
    {46x1 double}
    {31x1 double}
    {38x1 double}

xg2 = cellfun(@(g)x(g), xg2idx, 'Unif',0)                       % Option #2: ‘x’ Apportioned By ‘sort’ Indices

xg2 = 4x1 cell array
    {38x1 double}
    {46x1 double}
    {31x1 double}
    {38x1 double}

The first ooption just apportionns them as they exist in the original vector. Tthe second apportions them essentially by their percentile ranks in the vector by first apportioning the indices produced by the sort function.

I tried this with different lengths for ‘x’ and it appears to be robust. Obviiously there is a lower limit to the number of elements in ‘x’ that would probably crash it, however I didn’t do that experiment.

.

Answer 3

[Walter Roberson]

Perhaps use randsample

n = 4;
k = number of samples to generate
w = [0.25, 0.30, 0.20, 0.25];
y = randsample(n,k,true,w)