The output is now written into a netcdf file defined with the FastIsoProblem. This solves major memory problems when running high dimensions. This closes #115.
A reduced output to keep in the RAM can be defined through MinimalOutput, SparseOutput or IntermediateOutput. This is particularly helpful for docs, tests and inversion.
dudt is now converted to m/yr internally such that no floating point error occurs on examples like glacial cycles.
The inversion procedure has been generalised such that the user can define any workflow without having to go into the code. This closes #65 and #109.
The inversion is now threaded, which gives great performance improvement. This closes #64.
New layering options: "equalizing", "embedded" and "folded". This closes #30.
Correct an error in the computation of ice thickness above flotation. Previously using: z_ss + bsl. Now dz_ss + bsl.
Minors:
I have considered initializing arrays with undef but it produced some errors in some cases... Functionalities are there but are not used (zeros is preferred instead). This closes #116.
CUDA.jl had to be installed for docs and publication. This should fix docs and registering after that should solve #114.
moved solid earth densities to LayeredEarth since they can also be optimized for.
renamed seasurfaceheight and seasurfaceheight_perturbation to z_ss and dz_ss.
Majors:
FastIsoProblem. This solves major memory problems when running high dimensions. This closes #115.MinimalOutput,SparseOutputorIntermediateOutput. This is particularly helpful for docs, tests and inversion.dudtis now converted to m/yr internally such that no floating point error occurs on examples like glacial cycles.z_ss + bsl. Nowdz_ss + bsl.Minors:
undefbut it produced some errors in some cases... Functionalities are there but are not used (zerosis preferred instead). This closes #116.LayeredEarthsince they can also be optimized for.seasurfaceheightandseasurfaceheight_perturbationtoz_ssanddz_ss.