I’m starting to look through the source to assess the feasibility adding probability scales to bokeh.
For reference, here’s the introduction to prob scales in the docs for the maplotlib-based scale I’ve created:
http://phobson.github.io/mpl-probscale/tutorial/closer_look_at_viz.html#probability-plots
My plan is to start with an extension library at first (a la mpl-probscale), but if there’s interest in incorporating upstream, I’d love to do that too.
I’ve got just a couple of basic questions:
-
My initial assessment makes me think that the scales are mostly written in CoffeeScript . Is that correct?
-
If the scales are in CoffeeScript and probability scale needs the PPF and CDF functions of a normal distribution for the scale, those have to be implemented in CoffeeScript as well, right?
-
I see that the mantissas property of ticker classes control which values are ticked. How do you control the placement of the ticks? In other words, a probability scaled might be scaled [0.1, 0.2, 0.5, 1, 2, 5, … 95, 98, 99, 99.5, 99.8, 99.9], but those are placed based on the normal PPF of those probabilities (see link above). Where does that transformation of tick_value → tick_placement happen?
When creating my matplotlib scale, I used their LogScale as a template, but I’m having a little trouble following where the guts are in bokeh. A nudge in the right direction would be most appreciated.
-Paul