Bokeh backend

[rafajak]

Hi, Here are my scripts introducing a Bokeh backend.

Github won't render bokeh graphs, so here's a screenshot of example output: tiny.cc/qfs45y

I wasn't sure how to address plot sizing, as plot width and height in Bokeh are expressed in pixels, while matplotlib uses physical sizes and DPIs. I'm suggesting a change in livelossplot/generic_plot.py, which gives a default of figsize=[400,300] for each individual plot, if Bokeh backend is specified. Perhaps there are more elegant ways of approaching this - looking forward to hearing your thoughts on this :)

Cheers, Raf

[stared]

Thanks @rafajak ! Once improved, I would love to merge it.

A few comments:

• [ ] for drawing diagrams, don't clear_output(wait=True); it is possible to change a Bokeh blot in-place

Misc:

• [ ] Create a global target list for [BOKEH_TARGET, MATPLOTLIB_TARGET, NEPTUNE_TARGET]
• [ ] For screenshots, it would be wonderful to inline it in comments; I cannot access the one you sent.)