Matthew Badger

@noamross@ecoevo.social @rabaath@fosstodon.org @jimgar@hachyderm.io the distinction I’d draw between these and pandera is that while pandera does do these runtime checks, it also provides static type-checking when writing code in an IDE. It has its limits, but it’s super useful to be able to define a schema and the type checker *knows* that a dataframe has a particular column of a given type.

@rabaath@fosstodon.org as someone with more than a decade of experience with pandas, complaints about the api are common (and rightly so!); it’s a fantastic tool but a PITA much of the time. One minor point, if you do .groupby(…, as_index=False), it doesn’t set the group cols as the index. Another point: much of the fluidity the tidyverse is thanks to R vs Python, not it vs pandas. I’d love a top-level pipe in Python! (But R’s liberal approach to var declarations is problematic for GPPLs.)

@jimgar@hachyderm.io @rabaath@fosstodon.org so I was referring to non-standard evaluation and in particular the tidyverse’s data masking, which enables e.g. starwars %>% filter(homeworld == "Naboo").

It’s not an issue; it’s definitely nicer syntactically than Python, but it makes static type checking much more difficult. Not a problem for a REPL, but it makes it difficult to validate the correctness of a function/program/whatever.

(pandas now has Pandera, for type checking DFs, and I’ve found it very very useful.)