In pandas 0.25 we are decoupling the pandas visualization from the core pandas. This will make it possible to plot in a native way using third-party libraries. For example:
But at the same time, this will allow libraries replicating the pandas API to reuse the Matplotlib backend deployed with pandas, as well as the third-party backends.
While pandas has historically been implemented as a "batteries include" dataframe library, splitting some parts like the visualization, can help other libraries replicating the pandas API reuse components and not have to implement everything from scratch.
Are there other components that would be useful decoupling from the pandas core? Some candidates:
Extension arrays: Already "independent", but probably still too coupled to pandas internals to be reused?
In pandas 0.25 we are decoupling the pandas visualization from the core pandas. This will make it possible to plot in a native way using third-party libraries. For example:
But at the same time, this will allow libraries replicating the pandas API to reuse the Matplotlib backend deployed with pandas, as well as the third-party backends.
While pandas has historically been implemented as a "batteries include" dataframe library, splitting some parts like the visualization, can help other libraries replicating the pandas API reuse components and not have to implement everything from scratch.
Are there other components that would be useful decoupling from the pandas core? Some candidates:
pandas.indexing
Series.str
,Series.dt
,...pandas.plotting
: Not the backends (matplotlib, bokeh,...) but the code that creates the accessor and delegates the calls to the backend