topic Re: Correlation matrix to vector no diagonal and duplicates in SAS/IML Software and Matrix Computations

Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 12:56:18 GMT

Dear friends, consider correlation matrix (attached) and variables IBS_A, IBS_B and IBS_C. This is short example. I usually have much larger matrices with mostly unknown number of variables (goal is to have universal solution). I would like to obtain a vector "New" which would use data without diagonal and duplicates.

New

0,5

0,625

0,5

This is actually corelation between ab, ac, and ac. other is surplus. Can this be done?

Thanks!

Re: Correlation matrix to vector no diagonal and duplicates

PeterClemmensen — Mon, 14 Dec 2020 11:57:26 GMT

I think you forgot to post your question?

Re: Correlation matrix to vector no diagonal and duplicates

Ksharp — Mon, 14 Dec 2020 12:54:11 GMT

proc iml;
use sashelp.class;
read all var _num_ into x;
close;
corr=corr(x);
idx=loc(row(corr)>col(corr));
want=unique(corr[idx]);
print corr ,want;
quit;

Re: Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 12:58:29 GMT

Actually I started to write and in the process my mother called... Total mess 😉

Re: Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 13:15:34 GMT

Looks perfect!

Let me try it on real data.

Re: Correlation matrix to vector no diagonal and duplicates

Rick_SAS — Mon, 14 Dec 2020 13:19:33 GMT

See the article "Extract the lower triangular elements of a matrix," which defines a function called StrictLowerTriangular.

If you also want to get the pairs of variable names for each entry, see the code in this blog post about pairwise correlations.

Re: Correlation matrix to vector no diagonal and duplicates

IanWakeling — Mon, 14 Dec 2020 13:23:57 GMT

I think KSharp's code will work for you, but you may want to remove the call to UNIQUE as it will give you a vector with the correlation coefficients sorted into order - which you may not want. It will also remove duplicates if there are any.

Re: Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 13:32:48 GMT

Hi!

For the moment it looks OK.

I removed unwanted parts and i am checking on few files will I get all values.

Order is not important for now, because all I want to do later is to plot box plots from multiple matrices. This is actually genomic relationship matrix for one population and those values and its variance will be compared (via box plot) with another population.

So order is not important. Just unique pairs (a-c is fine, c-a is exactly the same, and diagonal must be one).

Re: Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 13:38:57 GMT

Hi!

Thanks!

I knew about StrictLowerTriangular but this "unique" part was an issue. It is easy with 5-10 animals but with 1000 and more...

Re: Correlation matrix to vector no diagonal and duplicates

MajaAGR — Mon, 14 Dec 2020 13:40:37 GMT

Works awesome!
You saved hours of my life.

Thanks!