Repository metrics
- Stars
- (1,818 stars)
- PR merge metrics
- (Avg merge 3d 8h) (5 merged PRs in 30d)
Description
The new modular structure with the 3 subpackages relies heavily on pip's bundles of optional dependencies. Conda doesn’t allow optional dependencies, but we can have extra packages to “compensate”. For example, matplotlib has matplotlib-base which has not backend and is recommended for downstreams to depend on and matplotlib which includes a recommended backend for users.
We currently have the following conda->pip equivalences
- arviz-base -> arviz-base
- arviz-stats -> arviz-stats[xarray]
- arviz-plots -> arviz-plots
We definitely need to update the dependencies of the already existing arviz package on conda forge. However I am not completely sure to what.
We could also add:
- arviz-stats-core -> arviz-stats
- arviz -> arviz[matplotlib]?
- arviz-plotly -> arviz[plotly] (most relevant for plotly because conda install arviz/arviz-plots plotly won’t work due to webcolors dependency missing)
- arviz-bokeh -> arviz[bokeh]
Having more feedstocks isn't a huge overhead but it takes some resources on conda forge's side and we'll also have to review their recipes every time we make a release even if the review is short.