Feedback before TableTransforms.jl v1.0

TableTransforms.jl is ready for a first major release and we would like to ask the community for feedback before moving forward with the registration of v1.0. The API is quite stable and we managed to add various transforms during the last month.

I would like to thank @eliascarv for his amazing work polishing type instability issues, adding laziness to Select/Reject transforms and many other features. Also, @jafar.isbarov for his first open-source contributions :heart:

We are happy to add more transforms to the list, review PRs and consider different design choices that do not affect the end-user interface. Contributions are very welcome.

7 Likes

Nice package!

apply and revert seem like equivalents of predict and reconstruct in StatsAPI. The latter has just been added so that MultivariateStats can implement it for things like PCA.

FWIW, StatsBase supports a few basic transformations but applied to matrices rather than tables, using standardize or fit with transform.

Cc: @wildart

1 Like

Thank you @nalimilan for the feedback :slight_smile: We adopted the apply/revert naming convention to move away from the statistics jargon since we plan to add different kinds of transforms. We also restricted ourselves to Tables.jl inputs because the lack of variable names with matrices makes the API low-level and cumbersome to work with in practice, e.g. dim=1 or dim=2 and names=[:a,:b,...] everywhere.