jubatus_ml.txt

Hi Jubatus guys,

We are proud to announce that Jubatus 0.9.0, our 8th major update, is now available.

In this release we have a plenty of new features and improvements:

- We released a new Python package called `jubakit`, a module to easily evaluate precision of models.
  See https://github.com/jubatus/jubakit for details.
- We now officially support Ubuntu 14.04 Trusty.
- Improved Nearest Neighbor engine performance up to 20x.
- New Bandit algorithm (Thompson sampling) is now available.
- New Classifier algorithm ("cosine" and "euclidean" k-NN) is now available.

The full changelogs are as follows:

===============================================

Jubatus (Core)
----------------------

* Improvements
    * nearest_neighbor: support multi-threaded nearest neighbor search (#260)
    * nearest_neighbor: improve random projection hash calculation performance (#246)
    * nearest_neighbor and recommender: improve bit_vector hamming distance calculation performance (#250)
    * classifier: add new recommender method (cosine/euclidean) (#234, #276)
    * bandit: add new bandit method (Thompson sampling) (#238)
    * classifier: improve get_labels API to return number of trained instances (#204, #272)
    * recommender: support unlearning in euclid_lsh algorithm (#189, #264)
    * Improve Travis CI test matrix (#263, #269)
    * Improve compatibility with FreeBSD 9 (#259)

* Bug fixes
    * recommender: remove unnecessary hash calculation in LSH/MinHash algorithm (#249, #273)
    * classifier: fix get_labels API does not return mixed labels in NN algorithm (#210, #271)
    * classifier: fix rows added via MIX does not trigger unlearner (#270)

Jubatus (Server)
------------------------

* New feature
    * jubakit <https://github.com/jubatus/jubakit>: Python module to easily evaluate precision of models (#1044)

* Changes to Supported Systems
    * Ubuntu 14.04 LTS is now officially supported (#1071)

* Improvements
    * classifier: improve get_labels API to return number of trained instances (#1109)
    * Add example configuration file for Nearest Neighbor (multi-threaded LSH) (#1111)
    * Add example configuration file for Classifier (cosine / euclidean) (#1108)
    * Add example configuration file for Recommender (euclid_lsh with unlearner) (#1108)
    * Configuration in model is used when loading model file on startup (#1081, #1107)
    * Improve compatibility with FreeBSD 9 (#1099)
    * Update manpages (#1112)
    * Code cleanup (#1102)

* Bug fixes
    * Fix jenerator test (#1103)

===============================================

We hope you enjoy this release!