Allow to sort the values of the dimensions used in a groupby operation

[maximlt]

Noticed while modernizing NYC Buildings (holoviz-topics/examples#386), it would be nice to be able to declare whether to sort or not the values of the dimensions used in a groupby operation.

import holoviews as hv
import pandas as pd

hv.extension('bokeh')

N = 12
df = pd.DataFrame({
    'cat1': ['A', 'A', 'A', 'A', 'C', 'C', 'C', 'C', 'B', 'B', 'B', 'B'],
    'cat2': ['y', 'y', 'x', 'x', 'y', 'y', 'x', 'x','y', 'y', 'x', 'x'],
    'x'  : range(N),
    'y'  : range(N),
})

hv.Points(df, ['x', 'y'], ['cat1', 'cat2']).groupby(['cat1', 'cat2'], dynamic=True)

Order is ACB, I'd like to sort to ABC.

[maximlt]

Ideally, this would need to be supported for both dynamic=True and dynamic=False. For the latter, in practice it seems that the values get sorted in most cases as HoloMap (the groupby container), like others, has sort=True by default.

Supporting sorting specific dimensions only is probably out of scope, at least that's not what I'm requesting here :)

[hoxbro]

I would not be opposed to have these just sorted by default. Though, this need to be updated in Panel, example of code:

diff --git a/panel/pane/holoviews.py b/panel/pane/holoviews.py
index a10fae643..e3a10aaa5 100644
--- a/panel/pane/holoviews.py
+++ b/panel/pane/holoviews.py
@@ -635,7 +635,11 @@ class HoloViews(Pane):
 
             vals = dim.values or values.get(dim, None)
             if vals is not None:
                 vals = list(unique_iterator(vals))
+                with suppress(Exception):
+                    vals = sorted(vals)
             dim_values[dim.name] = vals
             if widgets_type == 'scrubber':
                 if not vals:

[maximlt]

Though, this need to be updated in Panel, example of code:

Hm that'd work but I'm not sure how HoloViews would then let users toggle the sorting behavior.

When I looked at it, I thought about updating Dataset.groupby with a new kwarg (e.g. sort_dim_values). For dynamic=True this would be here:

holoviews/holoviews/core/data/__init__.py

Lines 993 to 994 in d586f3f

	dynamic_dims = [d.clone(values=list(self.interface.values(self, d.name, False)))
	for d in dimensions]

I didn't look much into the dynamic=False case. The dim values aren't obtained in Dataset.groupby but within each interface's groupby method. So sort_dim_values would need to percolate down these methods and be handled there. Note that the group container types often (always?) have a sort parameter (inherited from MultiDimensionalMapping, True by default) so there's possibly a way to leverage that. Final note, some groupby methods already have some code related to sorting, for instance Pandas' interface (for performance reasons?):

holoviews/holoviews/core/data/pandas.py

Lines 258 to 268 in d586f3f

	groupby_kwargs = {"sort": False}
	if PANDAS_GE_2_1_0:
	groupby_kwargs["observed"] = False
	data = [(k, group_type(v, **group_kwargs)) for k, v in
	dataset.data.groupby(group_by, **groupby_kwargs)]
	if issubclass(container_type, NdMapping):
	with item_check(False), sorted_context(False):
	return container_type(data, kdims=index_dims)
	else:
	return container_type(data)