Hi all,
This release includes a breaking change in the default configuration for ImplicitMF - the old default was to use rating values if they were present, which made it unnecessarily difficult to use in its normal configuration. 0.13 introduced an option and a warning about future default changes, and 0.14 changes that default.
I have also split several recommenders with external dependencies into their own packages. This includes implicit, HPF, and the TensorFlow recommenders. Releases of those packages are coming shortly (after I make sure they work with current dependencies). The HPF package will only be published on PyPI, because hpfrec isn't in Conda.
The other major highlight is that it's now a lot easier to configure the k-NN recommenders for implicit feedback mode, through a 'feedback' option.
Have fun, and let us know if you have any problems!
- Michael
--
I may send mail outside of working hours; I do not expect you to. He/him.