to_dict() and from_dict() functionality for Coregionalize Kernel and MixedNoise Likelihood class

[gehbiszumeis]

These are mandatory prerequites for tracking GPy.models.GPCoregionalizedRegression models with Machine learning lifecycle tools (e.g mlflow)

[ekalosak]

Looks like a CI error, otherwise LGTM. Any thoughts on the CI error?
Also, where're the from_dict implementations? Or, are they inherited?

[gehbiszumeis]

Looks like a CI error, otherwise LGTM. Any thoughts on the CI error?

Yes, according to AppVeyor it seems, that it fails already doing the build for almost all environments. It cannot install a package apparently

ERROR conda.core.link:_execute(502): An error occurred while installing package 'defaults::charset-normalizer-2.0.4-pyhd3eb1b0_0'.

No idea why, I didn't touch the requierements.txt. Only for the py3.5 environment it runs past this step and breaks when running the code coverage test

GPy: Saving user configuration file to C:\Users\appveyor\.config\GPy\user.cfg coverage run travis_tests.py ........S............S...............................................................................................................................................................................................................................................................................................................................S.............................................................................................Command exited with code -1073741819

Also, where're the from_dict implementations? Or, are they inherited?

Yes, the corresponding from_dict() methods of the super classes are used. I added only the _build_from_input_dict methods to add some specific parameters to the corresponding input_dict.

[ekalosak]

Fun enough, the errors are different across Python versions.

Python 3.5

## Python 3.6 thru 3.9

## Things to try - The `conda clean` command in the 3.6 error message - Looks like the 3.5 error code shows up [here](https://github.com/TryGhost/node-sqlite3/issues/1318) as well, something to think about...

[gehbiszumeis]

Things to try - The conda clean command in the 3.6 error message - Looks like the 3.5 error code shows up here as well, something to think about...

The 3.5 error disappeared, the >3.6 errors persist. I tried quite some stuff manipulating the way conda is trying to install packages, with no success yet. After all the trying it seems, that the charset-normalizer package has only a noarch version and no win64 version.

What really puzzles me is that the build fails only for this PR and not for others, so it cannot be some general bug in charset-normalizer or similar. It seems that for my build it tries to install that package, while for other PR builds it doesn't. That's weird

[ppk42]

I brought the appveyor build back to life. The pull request is still open. But, you'll find the changes in the last commit. (a93c4ac)

[codecov]

Codecov Report

Merging #951 (4286501) into devel (3e19a85) will decrease coverage by 0.08%.
The diff coverage is 30.30%.

@@            Coverage Diff             @@
##            devel     #951      +/-   ##
==========================================
- Coverage   54.59%   54.50%   -0.09%     
==========================================
  Files         211      210       -1     
  Lines       21564    21557       -7     
  Branches     2891     3168     +277     
==========================================
- Hits        11772    11750      -22     
- Misses       9240     9253      +13     
- Partials      552      554       +2     

[ekalosak]

Brilliant work both, thank you for the contributions!
Because the code coverage is still not really a priority, passing builds and tests is totally sufficient so I'm going to merge this.
@lawrennd @apaleyes <- please give this an 👁️ so we can remove it if it's bad.

[lawrennd]

Thanks @ekalosak and @ppk42!