Upload NUTS notebook

[I-Bouros]

Closes issue #28

[MichaelClerx]

Looks good to me but definitely one for @ben18785 :D

Few minor things:

• We can probably remove the line using subplots_adjust now, if that warning is to be believed?
• Is it worth going up to 20000 iterations like most other notebooks, for ease of comparison? (I thought at first that the KLD/ESS were low compared to other methods but that's not the case, just looks that way because of the lower iteration count!)
• Maybe switch off the screen output (controller.set_log_to_screen(False)) to make it easier to read? Similarly could replace [ecdf_norm_plotter(draws[:, i], np.sqrt(i + 1)) for i in range(6)] with 6 lines ecdf_norm_plotter(draws[:, 0], np.sqrt(1) etc so that the plots don't get grouped in a little scroll window?

@ben18785 do you know if we can easily compare to stan on any of these problems?

[ben18785]

Thanks. I think the number of iterations is fine -- NUTS is substantially more expensive than some other algos, so think we shouldn't aim for consistency on that.

I think otherwise all looks fine! (Not sure we need to go down comparison with Stan as it will likely differ a bit as they tweak their algo over time.)