Implementation of HCLUST-M (see interpolatedDistance())

[maxbox63]

New in this version:

• New function interpolatedDistance() implements an dissimilarity which penalizes (half the) squared Euclidean distance with penalty (1-a(r)) = 1 - FN(r) / (FN(r) + FP(r)), which directly mimics the adjacency function created for use in WCNA. For comparison, existing function betaDistance() penalizes (half the) squared Euclidean distance with the likelihood of the correlation coefficient (r) under the null model.
• Existing function estimateShape() has a different interpretation of the adjustment parameters "left" and "right". In the past these two parameters had an indirect effect on the final estimates; now they have the (expected) direct effect.
• New function estimateShape2() has adjustment parameters "middle" and "samples", which provide a more intuitive means of adjusting the estimated shape parameters. These are translated into "left" and "right" internally, and also have the expected direct effect.
• In the previous version there were actually three different conventions for specifying whether to interpret correlations as signed (correlation) or unsigned (association). All functions now use parameter signed=FALSE (default is unsigned).
• A few mistakes in the documentation of various parameters have been corrected.