Major refactor and cleanup

[mloubout]

Following has been done:

• Switch all vector types to proper abstract vectors. This should allow better generalization and interface with julia in general. From the structure of the problem, all data types are now multi-source vectors as Vector[Array} all based on a common judiMultiSourceVector type. To make it compatible with standard julia mat-vec the size of a judiMultiSourceVector is (nsrc,) but the length is the actual total length to match the operator size.

• Abstract operators better abstracted and implemented. Again removes a lot of code duplicate and makes it lot easier to maintain and use.

• Lots of cleanp of the wrapper functions that had a lot of duplicate. Now only misfit_fg for fwi/lsrtm objectives and time_modeling for all linear operators are left.

• Added additional tutorials from the ML4Seismic session

The user interface is mainly unchanged, the only breaking API changes are:

• Info been removed
• Sizes are now abstracted as symbols, this allow much cleaner management of the operator. Because info is gone, some of the size cannot be known at creation. THese "zero" sizes are updated to proper value when the linear operator is combined with another one making the propagation fully defined. The sizes can be converted to proper Intas well to be compatible with the rest of Julia (like IterativeSolvers that check all the actual length)
• judiLRWF (renamed judiWavelet but judiRHS still exist for backward compact) requires dt to be properly defined
• Most of "extended" stuff si removed and properly abstracted as a different projection and reuse same code

Can be discussed (at least in part) but some of it isn't very negociable and I'll be working with this version that is much easier to work with for external and dependent packages.

@philippwitte sorry, not much left of your original code :D

[philippwitte]

sorry, not much left of your original code :D Haha no problem.

Since there are breaking API changes, let's make sure to burn a release of the current version that people can fall back to if they don't want to update their code.

[mloubout]

let's make sure to burn a release of the current version

Oh yeah definitely, still got time need to add examples and double check stuff.

[codecov[bot]]

Codecov Report

Merging #100 (f1e7054) into master (e2307a4) will decrease coverage by 4.20%. The diff coverage is 70.99%.

@@            Coverage Diff             @@
##           master     #100      +/-   ##
==========================================
- Coverage   80.16%   75.95%   -4.21%     
==========================================
  Files          30       23       -7     
  Lines        2344     1801     -543     
==========================================
- Hits         1879     1368     -511     
+ Misses        465      433      -32     

Impacted Files	Coverage Δ
src/TimeModeling/Modeling/twri_objective.jl	0.00% <0.00%> (ø)
src/TimeModeling/Utils/seismic_preconditioners.jl	79.23% <ø> (ø)
src/TimeModeling/Utils/auxiliaryFunctions.jl	72.64% <42.59%> (-7.56%)	:arrow_down:
src/TimeModeling/Types/judiComposites.jl	76.64% <47.05%> (+0.04%)	:arrow_up:
src/TimeModeling/LinearOperators/lazy.jl	49.25% <49.25%> (ø)
src/TimeModeling/Modeling/distributed.jl	32.25% <57.14%> (-60.05%)	:arrow_down:
src/TimeModeling/Modeling/misfit_fg.jl	65.51% <65.51%> (ø)
src/JUDI.jl	83.33% <66.66%> (ø)
src/TimeModeling/Types/judiWavefield.jl	72.50% <70.27%> (-26.15%)	:arrow_down:
src/TimeModeling/Types/ModelStructure.jl	86.66% <71.42%> (-0.65%)	:arrow_down:
... and 21 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update e2307a4...f1e7054. Read the comment docs.

[mloubout]

Ok all comments fixed and added some cleanup and backward compatibility

[mloubout]

Any other comments

[mloubout]

Any other comments