Sorting categorical axes

[poplarShift]

Following up on holoviz/hvplot#122, I think we should first decide if ordering is a part of the dimension declaration or of the styling options.

One solution would be, first of all, to respect the ordering of the hv.Dimension(values= ) kwarg, that's not too hard I guess.

Beyond that, if we wanted to have ordering by, say, alphabetical, inverse alphabetical, length of label, whatever, or mean/max/etc, would that live in the styling options with a hook into redimensioning the values= kwarg accordingly? Or something different?

[philippjfr]

Respecting values should definitely be done, that is already outlined in #2468. Beyond that we definitely need some discussion and I wouldn't be opposed to some options to determine that. In the case of a categorical axis it may not be too weird to use xlim/ylim to set an explicit ordering for example.

[poplarShift]

Hm, I hadn't seen #3675. On the other hand I was sure I'd tested the functionality more recently than that and it still wasn't working in my case. Will have to investigate again.

[philippjfr]

On the other hand I was sure I'd tested the functionality more recently than that and it still wasn't working in my case

It's definitely not implemented yet.

[DancingQuanta]

I have this sorting issues with NdLayout/HoloMap not respecting the sort order from the original dataset in xarray.Dataset.

I did write a workaround for the meantime

def odict_sort_by_list(d, sort_list):
    return type(d)((k, d[k]) for k in sort_list) 
plot.data = odict_sort_by_list(plot.data, list)

where list is a list converted from a xr.coordinates and plot is either HoloMap or NdLayout.

[PeterFogh]

Problem solved - see the end of the message.

I have a similar problem of not being able re-order the axis ticks on a holoviews heatmap:

df = pd.DataFrame({
    'x': ['a', 'a', 'b', 'b'],
    'y': ['c', 'd', 'c', 'd'],
    'z': [1, 2, 3, 4]})
hv_heatmap = hv.HeatMap(df, kdims=['x', 'y'], vdims=['z'])
hv_heatmap

Have anyone been in the same situation? And perhaps found a workaround?

I have tried @DancingQuanta' workaround of overwriting the order of hv_heatmap.data, but without any luck.

I want to change the order of the y-axis, because I'm changing my current implementation of a confusion matrix from matplotlib to holoviews, such that it gets the same format as seen here: https://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html

EDIT
I solved the problem by using the invert_yaxis=True option flag (found it here: holoviz/datashader#596 (comment)).

df = pd.DataFrame({
    'x': ['a', 'a', 'b', 'b'],
    'y': ['c', 'd', 'c', 'd'],
    'z': [1, 2, 3, 4]})
hv_heatmap = hv.HeatMap(df, kdims=['x', 'y'], vdims=['z']).opts(invert_yaxis=True)
hv_heatmap

[MikeB2019x]

I'm having a similar issue as I explain in a question posted on stack. The solution isn't fixed by inverting axes though. Any suggestions?