Conda packaging

[stefanwezel]

• Make sbi installable through conda
• Based on existing PyPI

[codecov-commenter]

Codecov Report

Merging #516 (18f20e6) into main (df147b1) will not change coverage.
The diff coverage is n/a.

❗ Current head 18f20e6 differs from pull request most recent head fc4f754. Consider uploading reports for the commit fc4f754 to get more accurate results

@@           Coverage Diff           @@
##             main     #516   +/-   ##
=======================================
  Coverage   67.70%   67.70%           
=======================================
  Files          55       55           
  Lines        3970     3970           
=======================================
  Hits         2688     2688           
  Misses       1282     1282           

Flag	Coverage Δ
unittests	67.70% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out 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 df147b1...fc4f754. Read the comment docs.

[psteinb]

I couldn't find any documentation on this yaml quickly. I am a newbie to conda packaging.

But maybe this URL should point to our issues tab.

[stefanwezel]

This example from the conda docs suggests that it should point to the sbi GitHub repository itself.

[psteinb]

Ok, then there should be 2 changes according to this example from the conda docs.

Suggested change

	dev_url:
	dev_url: https://github.com/mackelab/sbi

And we have to decide whether the home key should have the same value or not. Not that it matters much. ;-)

[stefanwezel]

I would suggest

Suggested change

	home: https://github.com/mackelab/sbi
	home: https://www.mackelab.org/sbi/

[psteinb]

Sounds good, feel free to commit this.

[alvorithm]

OK, so it turns out we would need to recursively package the dependency tree with conda (as per conda/conda-build#548), which takes too much work now, so closing the issue until somebody finds time.

Note the current pip-only dependencies seem to be: pyknos (no further deps), nflows, which depends on torchtestcase (potentially more deps in turn) and UMNN (https://github.com/AWehenkel/UMNN) (unclear deps).