The Jubatus library is a online machine learning framework which runs in distributed environment.

See http://jubat.us for details.

LGPL 2.1

Improvements

• Asynchronous call to multiple servers at once, both keeper and mix - common/mprpc
• Refactor generator
• Error message improve
• Timeout in unittest expanded

Bugfix

• jubactl doesn't work ver 0.2 and later #13
• jubavisor doesn't work on daemon mode #5
• Asynchronous mprpc client critical bug
• #47, #50, #34, #36, #37, #31, #19, and other small bugfix

Improvements

• Simpler interfaces at classifier, regression and recommender
  • Clients are NOT COMPATIBLE with previous releases
• Now mix works concurrently in multiple threads (except tf-idf counting)
• Asynchronous RPC to multiple servers at once
• Add --version option
• Interface description language changed from C++-like to Annotated MessagePack-IDL
• Minor error handling
• A bit more tested than previous releases

Bugfix

• #30, #29, #22

Bugfix release: #28

New Features

• recommender
  • support fast similar item search, real-time update, distributed data management
  • inverted index : exact result, fast search
  • locality sensitive hash : approximate result, fast search, small working space
• regression
  • online SVR using passive agressive algorithm
  • as fast as current classifier
• stat
  • a Key(string)-Value(queue<double>)
  • O(1) cost of getting sum, standard deviation, max, min, statistic moments for each queue
• server framework
  • less-tightly coupled distributed processing framework with each ML implementation
  • idl & code generator - make it easy to write own jubatus system
  • removed public release of client libraries (so easy to generate!)
  • multiple mix - mutiple data objects can be mixed in one jubatus system

Bugfix

• duplicate key entry in fv_converter breaks the parameter

Bugfix release

Hello Jubatus!

First release: including classifier, and mix operation