Second release of the stat-fem code base. The release implements a number of improvements and fixes from the first release, including the following:
LinearSolver and other routines involving a solver can now accept keyword arguments to control the PETSc solver method and options. These are the same as the Firedrake options, so should enable straightforward integration between the two code bases.
A bug fix in the LinearSolver class has been implemented to ensure that the gradient of the log-posterior is correctly computed.
More transparency has been provided on how the mean of the solver output is scaled by providing a boolean scale_mean keyword argument that determines if the mean should be scaled by the model discrepancy scaling factor. By default, posterior solves do not scale the mean, while predictions do scale the mean, to maintain consistency with the previous behavior.
Improved documentation, including full docstrings for all functions and arguments, as well as a page on ensemble parallelism in the docs.
Second release of the
stat-femcode base. The release implements a number of improvements and fixes from the first release, including the following:LinearSolverand other routines involving a solver can now accept keyword arguments to control the PETSc solver method and options. These are the same as the Firedrake options, so should enable straightforward integration between the two code bases.LinearSolverclass has been implemented to ensure that the gradient of the log-posterior is correctly computed.scale_meankeyword argument that determines if the mean should be scaled by the model discrepancy scaling factor. By default, posterior solves do not scale the mean, while predictions do scale the mean, to maintain consistency with the previous behavior.