fix: properly treat Bar.base with stacked Bars

[maartenbreddels]

This also allows the use of bars with log scales

bug

We currently subtract base (baseValue in js code) each time from the value, which caused stacked bars to not show correct values. See demonstration of the bug:

Code to reproduce:

import bqplot
import numpy as np
import bqplot.pyplot as plt
N = 1000
np.random.seed(42)
log = False
sx = bqplot.LinearScale(min=0, max=4)
sy = bqplot.LinearScale(min=-2, max=6)
if log:
    sy = bqplot.LogScale(min=0.1, max=10)

fig = plt.figure(scale_x=sx, scale_y=sy)
x, y, c = np.random.normal(size=(3, N))
# mark = plt.scatter(x, y**2)
# h = plt.bar(x, y**2)
if log:
    mark = plt.bar([1,2,3], [[1, 1, 1], [2, 3, 4]], base=0.1, color=["orange", "green"], scales={'x': sx, 'y': sy})
else:
    mark = plt.bar([1,2,3], [[1, -1, 1], [2, -3, 4]], base=0, color=["orange", "green"], scales={'x': sx, 'y': sy})
plt.show()
import ipywidgets as widgets
if log:
    slider = widgets.FloatLogSlider(min=-1, max=1.5, step=0.01)
else:
    slider = widgets.FloatSlider(min=-1, max=7)
    
widgets.jslink((mark, 'base'), (slider, 'value'))
slider

solution

I've refactored the code a bit, which makes it easier to comprehend I think, and corrected the double subtraction.

Bars now behave as I expect:

There are a few issues/cornercase:

• If base is larger then the first bar value, there is no proper way to display it, so we skip it and print a warning to the console.
• Using a log scale, negative bars (i.e. where base is larger then the cumulate sum of y) makes totally no sense, but one could argue that's the user error.

[martinRenou]

Should the top of the bars actually be equal to base + y?

IIUC, that is what plotly does:

This would fix your issue with stacked bars, because bars would always have the same height, whatever the base is.

[maartenbreddels]

I think the current behaviour makes sense, given that if you plot in log space, and base is 10, and y value is 100, you want a bar from 10 to 100, not till 110

[ibdafna]

@maartenbreddels Are you still looking to pursue this? Looks not not based on the last comment, but I thought I'd ask 😃

[maartenbreddels]

Yeah, I think this is ready for review/merge. I just expressed I disagreed with Martin. Also, as it is now, the behaviour is the same as before, so non-breaking.

[maartenbreddels]

@mariobuikhuizen and I looked at this, we think this is an improvement, replaces #1375 and fixes #1300

Good to merge @martinRenou ?

[maartenbreddels]

It fixes a 'subset' of #1300, which is LogScale with non-negative values and a non-negative base (which the user has to set).

[martinRenou]

Thanks!

[maartenbreddels]

Thank you @martinRenou !

meeseeksdev backport to 0.12.x ?

[lumberbot-app]

Something went wrong ... Please have a look at my logs.

It seems that the branch you are trying to backport to does not exist.