Should `PearsonsCorrelation#computeCorrelationMatrix()` check NaN results?

[axkr]

Should PearsonsCorrelation#computeCorrelationMatrix() check NaN results here to detect errors earlier in the computation?

hipparchus/hipparchus-stat/src/main/java/org/hipparchus/stat/correlation/PearsonsCorrelation.java

Line 235 in 43dcbbd

double corr = correlation(matrix.getColumn(i), matrix.getColumn(j));

if (Double.isNaN(corr)){
  throw exception
}

Example matrix {{0.0,0.0},{0.0,0.0}}

[maisonobe]

When generating this exception, one of the reference test for MillerUpdatingRegression fails (the testPCorr test).
Looking at what this failing test does, it appears that the Pearson correlation matrix indeed as NaN entries in first row and first column (except diagonal element which is 1.0), but these entries are not used.
So I am not sure if we should throw the exception at this low level, as it prevents computing the rest of the matrix as soon as we get the first NaN.