debrouwere / datashell Goto Github PK

1. some packages should be available, but only namespaced – e.g. there's a ton of stuff in numpy that we don't want in the global namespace, but it's still useful to be able to use numpy
2. other packages export a ton but have only a small amount of very useful functions that should be in the global namespace, e.g. OLS in statsmodels.api
3. for still other packages, it does make sense to drop them into the global namespace (I'm thinking all of the functions from math, random and all of the scipy.stats.distributions) but alongside all of the good stuff they have a couple of useless or internal variables that are imported alongside when really that's not what we wanted – either a whitelist or a blacklist is needed
4. some packages have aliases that are so incredibly common that we should support them: np for numpy, pd for pandas

Like R, we should also warn when variables from one package mask those from another.

Like Django's runserver_plus, we should print out exactly what we import when we import it, both for clarity and to make it easier to move to production code.

Right now the CSVLoader is read-only. We'd need an interface that avoids surprise / "what the hell happened to my original data" though.