Add interpolator backends and MIR interface

[sandorkertesz]

This PR adds the following changes:

Backends

A Backend is an object implementing the interpolate() method. There are built-in backends:

"local-matrix": use a local matrix inventory from a user defined path. One object will be created per path. Paths can be added via the local-matrix-directories config option (set to None by default).
"remote-matrix": use a remote matrix inventory from a user defined url. One object will be created per url. Urls can be added via the remote-matrix-directories config option (set to None by default).
"system-matrix": the official remote matrix inventory with a hardcoded path
"mir": a MIR based interpolation. NOTE: this is not yet working!!!

Plugins

A backend can be implemented as a plugin.

Lazy object creation

An actual backend object is created lazily on first call of interpolate() on it.

Backend order config

When calling earthkit.regrid.interpolate() it tries all the available backends in a user specified order until one is able to perform the interpolation. The order is defined by the backends config option. By default it is set to None meaning that the following hardcoded default order is used:

DEFAULT_ORDER = ["local-matrix", "plugins", "remote-matrix", "system-matrix", "mir"]

E.g.

backends=None
backends="system-matrix"
backends="mir"
backends="system-matrix,mir"
backends=["system-matrix" ,"mir"]

Explicit backend order

The backend order can be directly passed to interpolate(). In this case the backends config option is ignored. E.g.

interpolate(vals, gs_in, gs_out, backends="system-matrix")
interpolate(vals, gs_in, gs_out, backends="mir")
interpolate(vals, gs_in, gs_out, backends="system-matrix,mir")
interpolate(vals, gs_in, gs_out, backends=["system-matrix", "mir"])

Breaking changes

The matrix_source keyword was removed from interpolate().

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 95.83333% with 1 line in your changes missing coverage. Please review.

Project coverage is 94.35%. Comparing base (389301b) to head (f0f3597).
Report is 1 commits behind head on develop.

Files with missing lines	Patch %	Lines
tests/test_memcache.py	83.33%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           develop      #60       +/-   ##
============================================
+ Coverage    52.55%   94.35%   +41.79%     
============================================
  Files           10       12        +2     
  Lines          588      885      +297     
  Branches        16       32       +16     
============================================
+ Hits           309      835      +526     
+ Misses         276       39      -237     
- Partials         3       11        +8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[pmaciel]

Is this mir-python? Or is this "from . import mir"?

[pmaciel]

I think "backend" is now "interpolator"

[pmaciel]

I think "backend" is now "interpolator"

[pmaciel]

I believe you've generalised the "method" into an "interpolator" which can be local or remote and so on, and interprets arbitrary arguments instead of specifically passing down directly options to mir. That's perfect! (Although, I cannot follow all of it)

In any case I think somewhere you have "import mir" which is a little confusing as theres a "mir.py" in the same directory, I think that should import the mir-python module in, but here it does a "local" mir.py import? Just a clarification

[sandorkertesz]

@pmaciel, thank you for the comments.

• Documentation has not been updated/written. Will be done when the "interpolator" concept is accepted
• The MirInterpolator will call MIR python whenever it will be available. The code and import there is just a placeholder and will be adjusted when the concrete MIR python method to call will be available.

[sandorkertesz]

method remained the same (e.g. "linear" etc.). Interpolator is something that executes the interpolation with the given options.

[pmaciel]

Ok thanks! So in my case that I'm using mir-python in my own branch, I should for the moment stay away from this one right? :-) (it's work on starting from gridspec)

[sandorkertesz]

Ok thanks! So in my case that I'm using mir-python in my own branch, I should for the moment stay away from this one right? :-) (it's work on starting from gridspec)

Yes! I have just changed the code in mir.py to better reflect what it should do in the future.