Hello Bokeh Users,

I searched for an answer to this question as I imagine someone else must have encountered it before. I have a pandas dataframe that I want to plot. I prefer to plot this data with “x_axis_type=‘datetime’”. However, the scatter radius bubbles aren’t plotted proportionally when I convert to datetime axis.

Here’s the plot with an integer indexed x-axis:

Here’s the plot with the same data except for x-axis is converted to a datetime axis:

You might be able to faintly notice the smallest specs of fill bubbles. They’re there but there small.

I tried to magnify the bubble size on the datetime index because I figured the datetime timestamp may be magnifying the proportions on the x-axis size. That didn’t work.

Can someone please assist?

Thanks,

Pete

I forgot to add my code snippets and I’ll include those below.. sorry about that…

Hello Bokeh Users,

I searched for an answer to this question as I imagine someone else must have encountered it before. I have a pandas dataframe that I want to plot. I prefer to plot this data with “x_axis_type=‘datetime’”. However, the scatter radius bubbles aren’t plotted proportionally when I convert to datetime axis.

Here’s the plot with an integer indexed x-axis:

est variable is my datetime list of dates for the x axis

esp variable is my list of y data

esq variable is my list of radius sizes

x_data = range(0, len(est))

p = figure()

p.scatter(x=x_data, y=esp, radius=esq, fill_alpha=0.6)

Here’s the plot with the same data except for x-axis is converted to a datetime axis:

est variable is my datetime list of dates for the x axis

esp variable is my list of y data

esq variable is my list of radius sizes

p = figure(x_axis_type=“datetime”)

p.scatter(x=est, y=esp, radius=esq, fill_alpha=0.6)

Hi,

This is due to the fact that the “data” aspect ratio is very far from 1-1, see e.g.

[https://github.com/bokeh/bokeh/issues/517](https://github.com/bokeh/bokeh/issues/517)

For some discussion. But basically boils down too: Circles are radially symmetric, but to actually draw them on the canvas, the radius has to be measured along some single dimension. By default, Bokeh uses the x-dimenson scale to map data distances to screen (pixel) distances for circles, which in your case results in the tiny circles (because the x-dimension is really “milliseconds since epoch” .

There are a couple of options:

• The direction to measure is configurable, you can have the circles scale their radii according to the y-dimension instead:

  http://bokeh.pydata.org/en/latest/docs/reference/models/markers/circle.html#bokeh.models.markers.Circle.radius_dimension

• You can use a size (which is in absolute pixel units) instead of a radius (which is in “data space” units)

• You can make the radii values much much larger, on the scale of the timestamp values (which are very large values)

Or, as an aside, consider a different approach to the visualization altogether. I would suggest in general that a bubble chart is not advised then whenever the scales of the x- and y- dimensions are not exactly identical.

Thanks,

Bryan

Thanks, Bryan.

I could use size and come up with some values in the following format:

dot_size = 4

ebq_max = np.max(eb.size)

ebt = list(eb.index)

ebp = list(eb.price)

ebq = list(eb.size / ebq_max * dot_size)

The ratio is locked to dot_size and the visualization stays the same even though I use different dot_size variables.

I may tinker around and come up with a better solution. I will post if I figure out a better way. Thanks for the great help thus far.