Creating a line plot with x_axis_type='log' fails when x_max < 1

[StevenCHowell]

When creating a plot with a logarithmic scale on the x-axis, if the maximum x-value is less than 1 the x-axis decreases from left to right, the scale is linear, not logarithmic, and the default view does not show the data. I tested various values for the maximum x-value and found that when x_max > 1, the x-axis increases left to right but the default view only shows x > 1. This code demonstrates the problem

import numpy as np
from bokeh.plotting import figure, output_file, show

output_file('test.html')

x = np.linspace(0, 0.95) # x decreases left to right, linear scale
# x = np.linspace(0, 1) # x increases left to right but default view shows, 1 < x < 1.05
# x = np.linspace(0, 10) # x increases left to right but default view shows, 1 < x < 10.5
y = np.sin(x)/x

p = figure(x_axis_type='log')
p.line(x, y)

show(p)

I tried declaring the x_range for the figure using x_range=(x.min(), x.max()) but this had to effect unless I removed the zero point from the data, i.e., x = x[x > 0]. With theses changes, the x-axis increases left to right and the default view shows the range indicated. Here is a functional workaround

import numpy as np
from bokeh.plotting import figure, output_file, show

output_file('test.html')

x = np.linspace(0, 0.95) # x decreases left to right, linear scale
x = x[x > 0]
y = np.sin(x)/x

p = figure(x_axis_type='log', x_range=(x.min(), x.max()))
p.line(x, y)

show(p)

I did not test all these same conditions with y_axis_type but I expect it to have the same problems. I did notice that the default view only shows y > 1, even if the data begins ay y < 1.

Here are the key points that need to be verified for logarithmic x and y axes:

• make sure the x-axis decreases left to right when the maximum x-value is less than 1
• make sure the y-axis increases bottom to top when the maximum y-value is less than 1
• adjust the default view to ignore x<=0 or y<=0 but show all other data (including x<1 and y<1)

[StevenCHowell]

Did this PR get into the 0.12.4dev8 build? If so, the axis is still rendered incorrectly if the maximum x-value is ~1 or less.

Here is a demo of what I am seeing:

import bokeh
print(bokeh.__version__)

import numpy as np

from bokeh.plotting import figure, show, output_notebook
from bokeh.layouts import gridplot
from bokeh.palettes import Colorblind8 as palette

output_notebook()

values = np.linspace(-1, 1, 21)
print(values)

size = 250
semilogx = figure(title='semilogx test', x_axis_type='log', width=size, height=size)
semilogy = figure(title='semilogy test', y_axis_type='log', width=size, height=size)
loglog = figure(title='loglog test', x_axis_type='log', y_axis_type='log', width=size, height=size)

semilogx.line(values[values != 0.0], values[values != 0.0])
semilogy.line(values, values) 
loglog.line(values, values)

fig = gridplot([[semilogx, semilogy, loglog]])
show(fig)

[StevenCHowell]

Just realized this PR likely did not get into the 0.12.4dev8 build (updated my comment to reflect that). I will run it from source with this example to test, then close this if it does work. Sorry I did not do that first.

[bryevdv]

This was not in dev8 I don't think. I want to make dev9 but am waiting to hear back from TravisCI support about an issue on their end causing one of our builds to continuously fail first.

[bryevdv]

@StevenCHowell from my testings: the original example now works. However, there seems to be problems with data values < 0 as best I can tell.

[bryevdv]

i.e. this works:

values = np.linspace(0.1, 0.9, 21)

But this is blank:

values = np.linspace(-0.1, 0.9, 21)

ping @clairetang6

Given that the OP code seems to work, I guess I'd be more in favor of a new issue at this point. But thoughts welcome

[bryevdv]

Screenshot for np.linspace(0.1, 0.9, 21):

screenshot for np.linspace(-0.1, 0.9, 21):

JS console reports "could not set initial ranges

[StevenCHowell]

This looks good. It can be a separate issue. Thanks @clairetang6 and @bryevdv, this was a big step in making bokeh easier to use for my data (sorry I never actually helped fix anything rather than only criticizing).