Uncertainty improvements

[dimalvovs]

The calculation of chi^2 which gets minimized during the mcmc sampling uses an estimate of uncertainty in the denominator. There is a problem with the current default estimation of uncertainty in dense matrix: the minimum value that the variance can take does not with data anyhow, being a constant 0.1. Let's imrove this!

1. prototype takes just one argument (in addition to data matrix) - the constant. Thus, the constant used both to scale both the scale data, and as a threshold of minimum value of uncertainty. -> Change to two parameters, both in description and in function implementation. For back compatibility default the 3rd param with the value of the 2nd.

   CoGAPS/src/math/MatrixMath.cpp

   Line 74 in 8a21f28

   Matrix gaps::pmax(Matrix mat, float p)

   CoGAPS/src/math/MatrixMath.h

   Line 27 in 8a21f28

   Matrix pmax(Matrix mat, float p);

2. use the new version of pmax in the uncertainty in the DenseNormalModel constructor. use the expectation of the atom size as the threshold (need to find and ensure it is calculated before uncertainty is initialized). optionally: atom size * threshold.

   CoGAPS/src/gibbs_sampler/DenseNormalModel.h

   Line 73 in 8a21f28

   mSMatrix(gaps::pmax(mDMatrix, 0.1f)),

[dimalvovs]

Looking here: CoGAPS.Rcheck/00_pkg_src/CoGAPS/src/gibbs_sampler/SingleThreadedGibbsSampler.h at line 139 (void SingleThreadedGibbsSampler::birth()) and float DenseNormalModel::lambda() const
{
return mLambda;
} in DenseNormalModule, let's set the threshold to DataModel::lambda or, the same, to mLambda. Sasha.