Function to load a .mat file (which can be inside or outside matlab folder) and perform analysis for finding constant or near-zero variance values

I have a matlab data file that has only one variable of type 1x1 struct which has 1x680 cell which has 680 MxN double values (values can be -ve, 0 or +ve). I want to load this data file using a function and find constant and near-zero variance values in this data set.

Thanks in advance.

Answer 1

[Walter Roberson]

[filename, pathname] = uigetfile('*.mat', 'Selected mat file');
if isnumeric(filename); return; end     %user cancel
filename = fullfile(pathname, filename);
datastruct = load(filename);
varnames = fieldnames(datastruct);
datacell = datastruct.(varnames{1});    %there should only _be_ one, but just in case there are more

variances = cellfun(@(M) var(M(:)), datacell);
has_small_variance = abs(variances) < 1e-5;
selected_cells = datacell(has_small_variance);

[HT]

Thanks Walter. I want to execute this as a function so that i can call it do the analysis. And the input to the function will be the mat file.

[Image Analyst]

Simply wrap it in a function. You should really learn how to write functions because it's essential for you to continue in MATLAB.

function selected_cells = GetCells(pathname, filename)
if isnumeric(filename); return; end     %user cancel
filename = fullfile(pathname, filename);
datastruct = load(filename);
varnames = fieldnames(datastruct);
datacell = datastruct.(varnames{1});    %there should only _be_ one, but just in case there are more

variances = cellfun(@(M) var(M(:)), datacell);
has_small_variance = abs(variances) < 1e-5;
selected_cells = datacell(has_small_variance);

Then, to call it:

selected_cells = GetCells('c:\some folder', 'whatever.mat')

You can rename GetCells to whatever you want.

So, why are you putting a double array into a cell, then putting the cell into a field of a structure? Can't you think of any other way to make it more complicated? Why not just set the field equal to the double array directly and not even use a cell? And even then, only if the structure is a structure array, otherwise just use the double array itself directly.

[HT]

It was just a cell of double values and not struct(I mentioned struct by mistake) . But when the file is saved in datastruct, it is saved as struct. Also, this is some data that I have already and not generated by me so i am mot trying to complicate anything here. Thanks.

[HT]

Can you explain what is happening in the line : variances = cellfun(@(M) var(M(:)), datacell);

[Walter Roberson]

var() by default works one column at a time, but you want the variance over the entire matrix within each cell, so we run a function on each cell that reshapes the content of the cell into a vector using the (:) operator, and then takes the variance of the resulting vector. That is a single numeric value for each cell entry so you do not need 'uniform', 0 in the cellfun call.

If you wanted to take the variance per column, then you would

cellfun(@var, datacell, 'uniform', 0)

but then you would need to define more clearly what it means for hte matrices to be similar in variance.

[HT]

Thanks.

Is it possible also to calculate other values, lets say mean of the cells in this code. And for specifying the threshold (variances<some value), is it possible to give control to user for specifying the value.

[Walter Roberson]

tol = input('What tolerance do you want?');

means = cellfun(@mean2, datacell);
variances = cellfun(@(M) var(M(:)), datacell);
has_small_variance = abs(variances) < tol;
selected_cells = datacell(has_small_variance);
selected_variances = variances(has_small_variance);
selected_means = means(has_small_variance);