How to count "gaps" consisting of designated values in a vector?

I have a very long vector of data that includes gaps of "invalid" values where data that's missing has been replaced with a designated missing data value (e.g. -1) or with "NaN".

invalid_values = [-1 NaN];
sample_data_vector = [22 23 22 24 -1 -1 -1 25 20 24 NaN Nan NaN 25 24 -1 -1 22 20 NaN 23];

I want to summarize info on the length of the gaps of invalid data (and no need to distinguish between whether the value is Nan or -1 or some other number). It might be helpful to know that my goal is to compare gap lengths over subsequent trials to determine whether gaps are getting longer/shorter and more/less frequent.

So for the above sample vector I'd get a matrix output that summarizes the following info about the gaps:

Length: 1 Number of occurences: 1

Length: 2 Number of occurences: 1

Length: 3 Number of occurences: 2

How would I write code to do this?

Answer 1

[Matt J]

One way, assuming you have the Image Processing Toolbox,

  bw=ismember(sample_data_vector, invalid_values)|isnan(sample_data_vector);
  S=regionprops(bw,'Area');
  gaplengths=[S.Area];
  
  H = histcounts(gaplengths,1:max(gaplengths)+1) %output histogram

[Al_G]

Hi Matt,

Thanks! I tried it this way and while the output summarizes info for where there's a value of -1, it seems to ignore NaNs. Is there a way to get it to take occurrences of NaN into account?

[Matt J]

I've fixed it.

[Al_G]

This works great! Thank you, Matt!

[Al_G]

Revisiting this a few years later in an attempt to improve my code. Is there a straightforward way to obtain the indices of where the gaps start and end?

The ideal output would look like:

gap.length = [3, 3, 2, 1];

gap.start_indices = [5, 11, 16, 20];

gap.end_indices = [7, 13, 17, 20];

[Matt J]

That will just be,

gap.start_indices = find( diff([0,bw])>0 );
gap.end_indices = find( diff([bw,0])<0 );

gap.length = gap.end_indices-gap.start_indices+1;