JSON support added and general cleanup

Lots of bugfixes, some minor refactoring (and runtime speedup!), and JSON is now fully supported throughout in addition to protobuf. JSON is now the default communication protocol between client and server, and you can compile in support for protobuf with "-DUSE_PROTOBUF=ON" added to your cmake statement.

Because of the above, we now rely only on the 3rd party libraries curl and libssl-dev for our testing suite test-accuracy. The core bemorehuman binaries don't rely on any 3rd-party libs now given the default installation.

What's Changed

• pre-calculate half of xy pair by @unbrand in #26
• 27 add support for json messaging by @unbrand in #28

Full Changelog: v1.2.0...v1.3.0

Custom http server now included

This release adds our own custom http server called hum. Why do that? Because we wanted to make it easier for people to get going with the software and not have to install a web server. With this release, hum is the default and if you want to use your own http server such as nginx, you just add "-DUSE_FCGI=ON" to the "cmake .." call during build-time. Please see the README for more details.

If you use hum, then you also don't need to install FastCGI. This is because while hum speaks standard http to the client process, hum uses its own protocol when talking to recgen. It's lighter and simpler than FastCGI. The end result is that hum is pretty fast. We estimate it's about 30% faster than nginx for typical bemorehuman workloads. However, we don't recommend using hum in production! hum currently caters to recgen's needs and only listens for POST requests. There's also very little error processing so hum is really best suited for development environments at the moment.

For this release, we also simplified the build process, updated the doc, and fixed lots of little bugs throughout the codebase.

What's Changed

• 8 organize build dirs by @unbrand in #17
• 19 add tiny webserver by @unbrand in #20

Full Changelog: v1.1.5...v1.2.0

Port to NetBSD

Main idea

Now we can work on NetBSD. This allows for running bemorehuman in a Unix environment and is a stepping stone for running bemorehuman on different architectures. The main idea is that bemorehuman can run in different kinds of environments. Some of these environments may be only periodically (or never) connected to the internet. These might be small self-contained NUCs or other devices. Such devices could be in supermarkets or other stores where it would interesting to offer recommendations to people. Think "Minority Report" where Tom Cruise's character is walking in a public place and he hears ads that are targeted to him. In the movie he is annoyed by the ads because it's clearly not interesting to him, but the technology to create a more relevant recommendation is here in bemorehuman.

v1.1.0

Main point

Version 1.1 introduces the ability for bemorehuman to take in events such as ratings, purchases, clicks, etc. and then generate new valences from them. All while staying live and serving clients. There is a minimal (about 2 seconds) amount of downtime when the auto-generated valence cache is swapped in -- requests are queued during this time.

This means the bemorehuman process is a lot more autonomous now and can update itself when presented with new input events.

What's Changed

• 2 enable event call by @unbrand in #11
• 1 auto update valences by @unbrand in #13

New Contributors

• @unbrand made their first contribution in #11

Full Changelog: v1.0.0...v1.1.0

Initial release

This is the initial release of the bemorehuman recommendation engine. Please test with local data to ensure it works well with your own data in your own environment before deploying to production.