multi_line and categorical axis

Hi all,

unless I am missing something, it looks like there might be some problem plotting multi_lines when an axis is categorical (bokeh version: 0.12.7).

Minimal setup to reproduce the problem in a Jupyter notebook:

from bokeh.plotting import figure, show
from bokeh.io import output_notebook
output_notebook()

factors = ["a", "b", "c"]
p = figure(plot_width=400, plot_height=400, y_range=factors)
p.multi_line(xs=[[1,2,3], [1,2,3]], ys=[["a","c","b"], ["b","a","c"]], line_width=2)
show(p)

Throws the following error:

Javascript error adding output!
TypeError: Cannot read property 'c' of undefined
See your browser Javascript console for more details.

While the following code, which is identical except for the fact that both axis are numerical, functions correctly:

p = figure(plot_width=400, plot_height=400)
p.multi_line(xs=[[1,2,3], [1,2,3]], ys=[[2,1,0], [3,2,2]], line_width=2)
show(p)

[bryevdv]

I do not quite understand what you meant by using numeric synthetic coordinates instead.

I mean that categories are mapped to numerical coordinates internally, and you could just pass those numbers in stead of the string factor names. Roughly speaking if you have categories [“a”, “b”, “c”] then those will map to (0.5, 1.5, 2.5), but things like range padding can affect the mapping, so some fine tuning may be required.

But there are lots of ways that might also work. Another option is simply don’t use real factors coordinates or a real categorical axis at all. Instead use numeric coordinates, with a regular axis, and change how the axis displays tick values by making a tick formatter. (i.e. to show different label names for specific tick locations). Or even simpler, using tick label overrides. This is probably what I’d actually suggest to try to start. Here is a simple example:

from bokeh.io import output_file, show
from bokeh.plotting import figure

output_file("test.html")

p = figure(x_range=(0,4))
p.xaxis.ticker = [1,2,3]
p.xaxis.major_label_overrides = {1: "foo", 2: "bar", 3: "baz"}

p.multi_line(xs=[[1,2,3], [1, 1, 2, 3]], ys=[[2, 3, 2], [4, 5, 5, 4]])

show(p)

As I said I simply don’t know how this could be fixed and still maintain support for nested categories. Nested categories were a crucial feature addition (so to be explicit and unambiguous: nested category support is more important than multi-line on categorical ranges), so this will just have to be an odd corner case to work around, I am afraid.

[bryevdv]

@robin-vjc by ambiguity,I mean this:

When you have a sequence of “nested categories”, say for grouped vbars, it looks like:

x = [ ["west", "sales"], ["west", marketing"], ["east", "sales"], ["east", marketing"]]

But that’s exactly the same form as for multiline. I am not sure there is any way to reconcile or disambiguate this. The handling of the property values and columns is at a very low level so it’s not something that could be handled at the level of MultiLine itself. Which doesn’t mean impossible, but does mean risky and expensive and for a single corner case with workarounds, probably not something that can be prioritized any time soon.