I disagree on the pyplot comments. My experience even is in the other direction lately. In the sense that I often find myself using pyplot as Plots.jl backend in the end, as it gave for those particular purposes the most visually pleasing results (such as this wall of pyplot produced loss-function plots: http://lossesjl.readthedocs.io/en/latest/losses/distance.html). My guess is that plotting is one of those topics where it is really easy to develop strong opinions on.