LogScale with multi_line

[kuk]

Hello!

I’m trying to plot data with a multi_line glyph and with the vertical axis being logarithmic.

The displayed plot is is autoscaled only on a small window of the available data. One can zoom out manually and see everything, but the default view eclipses some data. When using repeated calls for generating line glyphs, the vertical axis scales as expected. (To see that behaviour set the variable use_multiline to False) Also, when only one line is provided to the multi_line (reduce xs and ys to one element each), the range is selected correctly.

fig = figure(y_axis_type="log")
xs = [[0,1],[0,1],[0,1]]
ys = [[2,1],[100,200],[10,20]]

data = dict(xs= xs,
           ys= ys)

data_source = ColumnDataSource(data)

# Change this...
# ************************
use_multiline = True
# ************************
#

if use_multiline:
    fig.multi_line(xs="xs", ys="ys", source=data_source)
else:
    for x, y in zip(xs,ys):
        fig.line(x=x, y=y)

show(fig)

Is there a (simple) way to adjust the scaling with multi_line?

Follow-up/ Alternative:
In my project, I manage to get the Reset button to have the almost correct behaviour by adding a js_on_event(Reset, CustomJS…) that sets the y_range to start and end values determined manually.

js_for_reset = """
            logPlot.y_range.start = minStore.value;
            logPlot.y_range.end   = maxStore.value;
"""
self._log_plot.plot.js_on_event(Reset, CustomJS(args = dict(logPlot = self._log_plot.plot,
                                                                    minStore = self._minstorage,
                                                                    maxStore = self._maxstorage),
                                                   code=js_for_reset))

with self._(min/max)storage being invisible Spinners I just use to communicate the values from the Python side to the JS side, and Reset being bokeh.events.Reset.

However, even if I trigger the Reset call after the data-property ColumnDataSource has been changed, I still get to see the limited y_range. I guess that Bokeh does a autoscaling after all the calls related to the change of the data are done. Could I add a js_on_event call that is called after this autoscaling has been done? Is there an event like “the plot has been updated”?

[Bryan]

@kuk This just seems like a weird bug specifically between the interaction of auto-scaling and log axes and multi-lines. Please file a GitHub Issue so that it can be investigated and fixed. I am afraid I do not have any workarounds to suggest for now.

[_jm]

The problem exists in the current version of bokeh (2.3.2), and unfortunately it does not appear to be a recent regression; downgrading to 2.2.3 still shows the same behavior.

The following hack might suffice as a workaround. It essentially places invisible markers at the lower-left (min,min) and upper-right (max,max) coordinates of the data.

If your plots are dynamically changing, I am not certain of the implications, i.e. you might also need to move those markers accordingly if the rescaling does not behave as expected.

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
"""
import numpy as np

from bokeh.models import ColumnDataSource
from bokeh.plotting import figure
from bokeh.io import show

fig = figure(y_axis_type="log")
xs = [[0,1],[0,1],[0,1]]
ys = [[2,1],[100,200],[10,20]]

data = dict(xs= xs,
           ys= ys)

data_source = ColumnDataSource(data)

# Change this...
# ************************
use_multiline = True
# ************************
#

if use_multiline:
    fig.multi_line(xs="xs", ys="ys", source=data_source)
    
    xa = np.array(xs).flatten()
    ya = np.array(ys).flatten()
    fig.circle(x=[xa.min(),xa.max()], y=[ya.min(),ya.max()], alpha=0.00)
    
else:
    for x, y in zip(xs,ys):
        fig.line(x=x, y=y)

show(fig)

[kuk]

Thank you @Bryan and @_jm for the responses. I filed an issue on github, which apparently has been taken of already.

For now, your workaround works fine, @_jm. Thanks a lot!