Class FittedValuesForExcludedSamplesTest

java.lang.Object
ubic.gemma.core.util.math.linearmodels.FittedValuesForExcludedSamplesTest
public class FittedValuesForExcludedSamplesTest extends Object
Predicting a sample that did not contribute to the fit.

The sample-correlation matrix blanks flagged outliers before regressing, so they never influence the model — and their residual came back NaN, which is why the regressed matrix could never show what an outlier correlated at. The prediction was always computed (A.beta covers every row of the design) and then masked away; getFittedIncludingMissing() keeps it.

  • Constructor Details

    • FittedValuesForExcludedSamplesTest

      public FittedValuesForExcludedSamplesTest()

  • Method Details

    • aBlankedSampleIsPredictedButDoesNotInfluenceTheFit

      @Test public void aBlankedSampleIsPredictedButDoesNotInfluenceTheFit()
      Four samples, two groups, and the fourth blanked. The fit must be the fit of the other three, and the blanked one must still get a prediction.

    • theContributingSamplesKeepTheirResiduals

      @Test public void theContributingSamplesKeepTheirResiduals()
      The samples that did take part keep exactly the residuals they had. Filling in the excluded ones must not disturb anything else — that is what makes this safe to turn on for every dataset with a flagged assay.

    • withoutMissingValuesBothViewsAgree

      @Test public void withoutMissingValuesBothViewsAgree()
      With nothing missing the two views are the same matrix of numbers, so no caller can be surprised by reaching for the new one.