GitLab

The caching of the fitted models is quite compicated now. It raises the problems in multiprocessing mode, as well as potential memory leaks.

Ideas:

• move cache from Chain/Node to the external component, that allow preserving chain fitting results and can be passed as chain.fit(data, cache)
• save cache to disk instead of in-memory representation (e.g. use the pickledb)

May be @YanaPolonskaya also has some propositions as a part of PR #215.