Stop loop when first result is hit / loop of comparison bewteen different tbl heights?

My question can be formulated in 2 ways because I'm not sure about the strategy to use.

I'm a beginner coder, and have to make a comparison between two pieces of tables (output of a bigger code) that have different heights. I need to see if the start and stop days reported are overlapping between these tables. Those are days of assumption of two different drugs on a single patient, I need to see when they were assumed in the same days, but of course neither was given for the same amount of time, therefore tables have a different number of rows.

My idea was to report in the results table a 1 when there was one overlap in the days, and 0 if there's none. I don't care how many times they overlap, I only want to see IF there's an overlap and report a 1.

drug1_data:

PatientID    StartDay    StopDay      
{'1008'}        -2          0       
{'1008'}         2          5       
{'1008'}         9          9       
{'1008'}        12         13       
{'1008'}        16         16       
{'1008'}        19         20 

drug2_data:

PatientID    StartDay    StopDay      
{'1008'}         1         11
{'1008'}        13         17

My piece of code:

 if size(drug1_data, 1) > 0 && size(drug2_data,1) > 0
            % if drug 2 was stopped before drug 1 started or vice versa, there's no overlap
            % so if this is not the case, there is an overlap and we store a 1 in results 
            if ~(lt(drug1_data.StopDay, drug2_data.StartDay) || lt(drug2_data.StopDay, drug1_data.StartDay))
                results(i,j) = 1;
            end
 end 

But the tables have different heights and I get the obvious error: "Matrix dimensions must agree"

(i and j come from previous loops, and results is built as

results = zeros(length(unique_patients), length(DRUGS))

where DRUGS is the array of different drugs for the overlap comparison)

Giving that I'll use this code for hundreds of samples and they'll have all different heights all the time, how can I overcome this issue once and for all??

Thank you!

Answer 1

[Walter Roberson]

overlaps = ...
    (drug1_data.StartDay <= drug2_data.StartDay.' & drug2_data.StopDay.' <= drug1_data.StopDay) | ...
    (drug2_data.StartDay.' <= drug1_data.StartDay & drug1_data.StopDay <= drug2_data.StopDay.');

This will be an array that is height(drug1_data) by height(drug2_data) . In this case, 6 x 2, for the 6 courses of drug 1 and the 2 courses of drug 2.

any(overlaps, 1) will be a vector of length height(drug2_data) that is true where at least one course of drug1 overlaps with a course of drug2 (that is, will tell you which courses of drug 2 are overlapped). any(overlaps, 2) will be a vector of length height(drug1_data) that is true where at least one course of drug2 overlaps with course of drug1 (that is, will tell you which courses of drug 1 are overlapped). any(overlaps(:)) will be true if there is some overlap between drug 1 and drug 2.

[camcecc10]

What is drug2_data.StartDay.' for? I mean, the .'

[Walter Roberson]

When you extract from the table, you get two column vectors of different lengths. You want to compare each one of the first against each one of the second.

There are multiple ways to compare each of the first against each of the second. One way would be to use ndgrid or meshgrid to create explicit grids of data, like what you would get from repmat() each to have as many copies as needed to match up,

repmat(A, 1, length(B))
compare to
repmat(B, 1, length(A)) transposed

a1 a1
a2 a2
a3 a3

compared to

b1 b2
b1 b2
b1 b2

ndgrid and meshgrid can create those two forms automatically given the vectors involved.

This approach involves making explicit copies of the values.

Another approach is to use an older function that runs comparisons without building the explicit arrays

bsxfun(@ge, A, B.')

This processes a column at a time and puts the result together automatically.

The third approach, available in new enough matlab releases, is implicit expansion. When you use a binary operator between a row vector and a column vector, matlab will now know to automatically do expansion to operate on each member of the first together with each member of the second, same as if you had used bsxfun, which in turn is the same effect as if you had created grids of values (but without needing the memory involved)

So what the .' operation is doing in my code is flipping the second vector from being a column vector to being a row vector. The first vector is left as a column vector. Now you have column vector comparison row vector and matlab does the implicit expansion, comparing each member of the first against each member of the second.

[camcecc10]

Your explanation was just perfect, you taught me a lot in few lines. Thank you for both this useful comment and the answer, it's exactely what I was looking for for ending my code!

[Walter Roberson]

Which of the any() calls ended up being the one that was useful for you? Your original question suggested any(overlaps(:)) but I suspected that you might need the other information at some point.

[camcecc10]

any(overlaps(:)) was the one, in the end I need only one result for each patient