Better control over data/screen aspect ratios needed for large circles

UPDATE SUMMARY

Te underlying browser API accepts one radius parameter (in pixels). If data/screen aspect ratios are not matched, then the the data-space radius in different dimensions will not match, because “10” in data units will be a different number of pixels in the x- vs y-dimensions. This can also cause issues for hit testing. Things that can be done:

• using something that scales in both dimensions (e.g. p.ellipse instead of circles)

• make the aspect ratios match up by controlling padding, labels size and orientation, and canvas size and data ranges explicitly.

Making the aspect is a slight chore currently, but still doable. There is an open issue to make controlling aspects ratios easier to do.

In the event different aspect ratios are necessary, then this situation is mathematically unavoidable with canvas circles, and so they may simply not be appropriate to use in that case (i.e. use ellipses instead)

---

As far as I can see, the issue is caused by a large (read: orders of magnitude) scale difference between the x and y coordinates. Here’s an example, modified from hover.py, that triggers the issue in 0.4.2:

from __future__ import division

import numpy as np
from six.moves import zip
from collections import OrderedDict
from bokeh.plotting import *
from bokeh.objects import HoverTool, Range1d

TOOLS="pan,wheel_zoom,box_zoom,reset,hover,previewsave"

x = range(0, 1000, 10)
y = range(10000,20000,100) # <--

radii = [5]*100
colors = ['black']*100

source = ColumnDataSource(
    data=dict(
        x=x,
        y=y,
        radius=radii,
        colors=colors,
    )
)

output_file("hover.html")

hold()

circle('x', 'y', radius='radius', source=source, tools=TOOLS,
       fill_color='colors', fill_alpha=0.6,
       line_color=None, Title="Hoverful Scatter")


# We want to add some fields for the hover tool to interrogate, but first we
# have to get ahold of the tool. This will be made easier in future releases.
hover = [t for t in curplot().tools if isinstance(t, HoverTool)][0]

hover.tooltips = OrderedDict([
    ("index", "$index"),
    ("(x,y)", "($x, $y)"),
    ("radius", "@radius"),
])

show()  # open a browser

[bryevdv]

@leopd The developer’s guide is here:

http://bokeh.pydata.org/en/latest/docs/dev_guide.html

Please feel free to help out, some of us are already working at our limits.

[bryevdv]

Explicitly for everyone who needs to see it: #6784 added a match_aspect option that can be set to True when needed for “large” circles.