Bar plots with LogScale don't show up properly

[sntgluca]

Hi,

Describe the bug
I can't make Bar plots work using a LogScale scale.
I imagine the problem must be connected with the need for the minimum value of a LogScale be non-negative. It's similar but not the same as #1219.

To Reproduce
Here's a simple working example using a LinearScale:

import bqplot as bq
x = [i*3 for i in list("abc")] 
y = np.abs(np.random.randn(10) * 100)

sc_x = bq.OrdinalScale()
sc_y = bq.LinearScale()
scales = {"x": sc_x, "y": sc_y}

line = bq.Lines(x=x, y=y, scales=scales, marker='circle')
bars = bq.Bars(x=x, y=y, scales=scales, opacities=[0.1])

ax_x = bq.Axis(scale=sc_x, grid_lines="none")
ax_y = bq.Axis(scale=sc_y, grid_lines="none", orientation="vertical")
bq.Figure(marks=[line, bars], axes=[ax_x, ax_y], title="Linear scale")

This example doesn't work at all:

sc_x = bq.OrdinalScale()
sc_y = bq.LogScale()
scales = {"x": sc_x, "y": sc_y}

line = bq.Lines(x=x, y=y, scales=scales, marker='circle')
bars = bq.Bars(x=x, y=y, scales=scales, opacities=[0.1])

ax_x = bq.Axis(scale=sc_x, grid_lines="none")
ax_y = bq.Axis(scale=sc_y, grid_lines="none", orientation="vertical")
bq.Figure(marks=[line, bars], axes=[ax_x, ax_y], title="Empty log scale")

And setting the minimum value for the LogScale leads to weird behaviour:

sc_x = bq.OrdinalScale()
sc_y = bq.LogScale(min=0.1)
scales = {"x": sc_x, "y": sc_y}

line = bq.Lines(x=x, y=y, scales=scales, marker='circle')
bars = bq.Bars(x=x, y=y, scales=scales, opacities=[0.1])

ax_x = bq.Axis(scale=sc_x, grid_lines="none")
ax_y = bq.Axis(scale=sc_y, grid_lines="none", orientation="vertical")
bq.Figure(marks=[line, bars], axes=[ax_x, ax_y], title="Log scale")

Expected behavior
I would expect that Bar plots should be possible using a LogScale, and maybe return an error if the dataset doesn't allow it (if values are non-positive).

Additional context
Tested on bqplot 0.12.20.

Thanks!

[sntgluca]

Dear all,

Is there anyone who could take a look and give us feedback about this and #1219?

As a user who can't really help with the development, I don't have intention to put pressure but I think the use cases in these tickets are pretty important to be addressed.

[sntgluca]

Hi,
I was able to make it work in some specific case using type="grouped", and setting base as positive.
The fact that it doesn't work with type="stacked" makes me think that corrections to the base code are necessary, though.

x = [i * 3 for i in list("abc")]
y = np.abs(np.random.randn(10) * 100)

sc_x = bq.OrdinalScale()
sc_y = bq.LogScale()
scales = {"x": sc_x, "y": sc_y}

bars = bq.Bars(x=x, y=y, scales=scales, opacities=[0.1], type="grouped", base=1)

ax_x = bq.Axis(scale=sc_x, grid_lines="none")
ax_y = bq.Axis(scale=sc_y, grid_lines="none", orientation="vertical")
bq.Figure(marks=[bars], axes=[ax_x, ax_y], title="Linear scale")

[maartenbreddels]

I think #1423 fixes this, although I will only work for non-negative values and a non-negative base (see #1423 (comment) )