GitHub

Input conversions

Our plotting functions forward their documented arguments to the underlying Python functions, where necessary transparently converting acceptable Julia types to the following corresponding Python types.

DataTree arguments

Wherever a Python function accepts a DataTree, pass one of:

These are transparently converted to a Python xarray.DataTree.

Behavior change in ArviZPythonPlots 0.2

Pre-1.0 arviz used to auto-convert other inputs internally — bare numeric arrays, NamedTuples of arrays, raw Dicts of arrays. If you have one of these looser types, convert it to one of the above DataTree-like types explicitly first, e.g. with InferenceObjects.convert_to_inference_data.

dict of {str: DataTree} arguments (multi-model comparison)

Functions whose docstring also documents a dict-of-DataTree form (e.g. plot_dist, plot_forest, combine_plots) additionally accept an AbstractDict whose values are each one of the DataTree-like types above; the keys become model names. plot_forest/plot_ridge also accept a plain Vector/Tuple of such values, naming them positionally ("model1", "model2", ...).

DataArray arguments

Wherever a docstring documents an argument as a raw xarray.DataArray (e.g. plot_pair_focus's focus_var), pass a DimensionalData.AbstractDimArray; its dimension names and coordinates are preserved in the conversion.

Result-object arguments

A few functions take a specific Python type produced by another arviz function rather than a DataTree:

Everything else

Arguments not covered above (var_names, coords, backend, etc.) are converted generically:

  • Symbols become Python strs.
  • NamedTuple/Dict become Python dicts.
  • Vector/Tuple become Python list/tuples.
  • numeric arrays become NumPy arrays.

A PythonCall.Py object is always passed through unchanged, so you can build an argument with arviz/arviz_plots/xarray Python calls directly if needed.