import openml
openml.config.get_cache_directory()   # is:   '~\\.openml\\org\\openml\\www'
openml.config.cache_directory         # is:   '~\\.openml'

get_cache_directory gets the cache for the currently configured server, while cache_directory gets the root of the cache directory.
set_cache_directory(str) configures the root.

If we plan to maintain an external interface for both types, I propose we:

• we make them both functions so they are accessed in the same way (and using a function will make a get\set pattern easier where a set can be optionally for the session or saved to the configuration file)
• make it clear in the name what (part of) the cache directory is returned