Skip to content

[WIP] Covariance matrices

Emanuele Roberto Nocera requested to merge detatch_exp_covmat into master

Created by: nhartland

Here is my current WIP attempt to resolve #228 (closed) and to make solving #25 (closed) easier.

The steps I have in mind are

  1. Write a 'low-level' version of ComputeCovMat from chisquared.cc that operates on the basic quantities (systematics, t0/data vectors, statistical errors). This is called ComputeCovMat_basic and sidesteps (for now) the discussion in e.g #228 (closed) as to what the basic types should be.

  2. Rewrite both ComputeCovMat and Experiment::GenCovMat to use ComputeCovMat_basic, thus unifying the method used to compute Dataset and Experiment covariance matrices.

  3. Use ComputeCovMat_basic to generate sampling matrices, and use these for artificial replica generation.

------ Done up to here

  1. Have validphys use ComputeCovMat_basic (or some suitable wrapper) to compute the experimental covariance matrix.

  2. Remove the covmat generation from the Experiment class and replace it with read-from-file ones (I guess this is PR #244).

At that point we would have

  • All dependencies on fSys structure etc focussed in one place (ComputeCovMat_basic).
  • An external function that can be called from validphys to generate the experimental covariance matrix

Merge request reports