I was trying to figure out how to build a boxplot with Bokeh today. Obviously I want a bunch of rectangles at [(x,0),(x+width, y)]. And indeed, that is what http://bokeh.pydata.org/docs/gallery/histogram.html does. But it isn’t obvious.
Looking at the boxplot example, it’s very cool that you can compose renderers with bokeh to create complex plots. http://bokeh.pydata.org/docs/gallery/boxplot.html . I would expect a higher level boxplot interface.
I created a multiplot interface in 20 lines of python code here. http://nbviewer.ipython.org/github/paddymul/bokeh_tutorial/blob/master/multiplot.ipynb
That builds an interface to assemble multiple plots with a linked x-axis. We should have more examples like that in documentation.
I think we should have more guidance as to when an abstraction should be built into bokeh (as a pull request?). When it should be layered on top of plotting.py in application code. When are JS level mods required?
Any thoughts?
I specifically want a bar-chart interface built into the js api.
I think the glyphspec should look like this
{‘type’:‘barchart’,
‘x’:‘x’,
‘y’:‘y’,
‘width’:5,
‘width_units’:‘screen’}
Actually how should the width units be handled? does something intelligent need to be done for plots with a lot of datapoints.
My above glyphspec is frustrating because it would require me to write a renderer, when I don’t really want a renderer, I want a chance to put in some extension code.
Any more thoughts?
Paddy
