Add stacked lines to stacked area plot

Is your feature request related to a problem? Please describe. I wonder whether stacked lines could be added to stacked area plots (vstack_area()), which is the default in Seaborn or Plotly for example.

Describe the solution you’d like Stacked line (vline_stack()) and area (vstack_area()) plots take mostly the same arguments. One solution would be to just call vline_stack every time vstack_area is called. The solution I wanted to suggest in a PR for discussion is a bit more flexible, as it allows a line_width argument to be specified, in which case vline_stack is called. bokeh/plotting/figure.py:

def varea_stack(self, stackers, **kw):
    result = []
    if 'line_width' in kw:
        line_width = kw['line_width']
        kw.pop('line_width')
        result.append(self.vline_stack(stackers=stackers, line_width=line_width, **kw))
    for kw in double_stack(stackers, "y1", "y2", **kw):
        result.append(self.varea(**kw))
    return result

instead of

def varea_stack(self, stackers, **kw):
    result = []
    for kw in double_stack(stackers, "y1", "y2", **kw):
        result.append(self.varea(**kw))
    return result

This relies on vline_stack() and vstack_area() not diverging in future versions however. You will also notice that the legend has changed: it can be fixed if people think it’s an issue.

Describe alternatives you’ve considered One could just call vline_stack() whenever they make a stacked area plot, but other libraries seem to have chosen a different default.

Additional context Without stacked lines

np.random.seed(seed=10)
df = pd.DataFrame(np.random.randint(10, 100, size=(15, 10))).add_prefix('y')

p = figure(x_range=(0, len(df)-1), y_range=(0, 800))
names = ["y%d" % i for i in range(10)]

p.varea_stack(
    stackers=names, x='index', 
    color=brewer['Spectral'][10], 
    legend_label=names, 
    source=df,
    line_width=2, 
    alpha=0.5,
)
p.legend.items.reverse()

show(p)

With stacked lines

[bryevdv]

@harabat Offhand I think I am -1 on adding line properties to the low level VArea itself. I think your initial implementation, that confines changes to the higher level “stacked” glyph helpers, is a perfectly reasonable direction to pursue, it just needs to be more careful and thorough about handling the arguments. I’ll have to think about it some more.

Regardless, I am definitely reticent to start defaulting the alpha to anything other than 1. There is no default glyph in Bokeh that this is the case for, and apart from being a breaking change, I think it is good to keep the consistent expectation “If I want alpha, I have to set it” true everywhere. I’d be more inclined to follow the seaborn example where the line and area alpha are both 1 but the line color defaults to white to leave “line gaps” between the areas. We would want to start off with a default line_color=None in 2.x to avoid breaking changes, then change to line_color="white" default in Bokeh 3.0 when a breaking change can be made.

[harabat]

Turns out I forgot we had identified several options and didn’t actually decide on either.

Rereading through the discussions, I’m more inclined to go with my original approach and your comments on it: confine changes to varea_stack, while making sure all lineprops arguments are accounted for, and add line renderers with line_color=None by default (to switch to line_color='white' in 3.0).

My gripes with the option of adding a new vstack helper are that it’d be a confusing stack function, compared with the more clear cut varea_stack, vline_stack, and vbar_stack. Also, I can’t help but side with Seaborn and Pandas on the idea of having lines by default in stacked area charts.

Obviously, the above holds for the horizontal equivalents as well.