Extending Dataframe

[paddymul]

I'm interested if there are recommended best practices for extending DataFrames.

I work on a project where we have an API for querying a database that returns DataFrames. I basically want a way to associate extra metadata with each column in the DataFrame. At a minimum this would allow me to run commands like df.some_column.canonical_url and get a link back to a web service.

There are some other things that I would like to do like giving columns aliases for better legends when plotted.

I know I could return a wrapped DataFrame object from the api, but the DataFrame would lose its wrapper once operations were performed.

[jreback]

see #2485 for a long discussion.

You can do this in >= 0.14.0 by utilizing __finalize__().

see this PR's tests for basically doing this. https://github.com/pydata/pandas/pull/6931/files

Its straightforward, but you have to deal with the propogation issues. E.g. How to decide between different meta data.

[jreback]

closing as discussion in #2485