Slim derivatives

[arokem]

For some large-scale datasets (and just in general?) it may be good to think about ways to slim down the derivatives generated by pyAFQ. For example, we might want a minimal set of derivatives to be stored (e.g., export("just_profiles"), which only stores the final tract profiles csv and dumps everything en-route, especially things imported from other pipelines (to avoid duplications).

Another thing to consider is whether compressing some derivatives may be a good approach. For example, the html visualization can be compressed by a substantial amount (in a quick experiment from 29.3MB to 9.4MB in one case), so maybe zipping it up as a first pass may make sense under some conditions.