ChangeLog.rst

Release 1.1.1 - 2019/01/28

• Improvements

  • Suppress warnings at compile (#400, 401)

Release 1.1.0 - 2018/08/27

• Improvements

  • Use double instead of float (#397, 399)

Release 1.0.9 - 2018/04/23

• New Features

  • Clustering: Add distance metrics option (#339, #396)

Release 1.0.8 - 2018/02/26

• Improvements

  • Anomaly: Use reference in add_bulk (#394)

Release 1.0.7 - 2017/12/18

• New Features

  • Anomaly: Add bulk API (#360, #391)

• Bug fixes

  • Unlearner does not be saved (#389, 392)
  • Some unittests do not run on Python 3 (#390, #393)

Release 1.0.6 - 2017/10/30

• New Features

  • Recommender: Add filtering API (#387)

• Bug fixes

  • Failed to build on MacOS High Sierra (#388)

Release 1.0.5 - 2017/08/28

• Improvements

  • bandit: add epsilon-decreasing method (#354, #381)
  • Improve compatibility with gcc-6.x (#368, #369)
  • Improve compatibility with FreeBSD (#376, #384)
  • Improve compatibility with environment missing ld.gold (#371)
  • Implement equality comparator for json objects (#380)
  • Improve string to number conversion (#377)

• Bug fixes

  • nearest_neighbor: fix crash when search neighbors for empty model (#382, #383)
  • Fix FMV not working (#378, #379)

Release 1.0.4 - 2017/06/26

• Improvements

  • bandit: add seed option (#363, #365)
  • anomaly: imporve get_status to return additional metrics (#372)

• Bug fixes

  • Fix build issues on some platforms (#349, #373)
  • Fix --disable-eigen build option not working (#361, #362, #370)

Release 1.0.3 - 2017/04/24

• Improvements

  • regression: add distance-based weighting option (#351, #352)
  • anomaly: improve to avoid causing inf (#348, #358)
  • recommender: improve to avoid causing scores larger than 1.0 in inverted_index (#355, #356)
  • Improve local_storage model serialization format be compatible with local_storage_mixture (#359)

Release 1.0.2 - 2017/02/27

• Improvements

  • Support building on Ubuntu 16.04 (#282, #345, #346, #347)

• Bug fixes

  • Fix double acquision of read locks (#342)

Release 1.0.1 - 2016/12/26

• Improvements

  • fv_converter plugin loading code moved from Jubatus server (#338)
  • Add rwmutex sanity check assertion (#312)
  • Upgrade waf to support build with Python 3.5 (#277, #308)

• Bug fixes

  • anomaly: fix "ignore_kth_same_point": false treated as true (#340)
  • classifier: fix euclidean distance classifier name (#341)
  • classifier: fix unlearner not protected by mutex (#311, #343)

Release 1.0.0 - 2016/10/31

• New Machine Learning Algorithms

  • clustering: add DBSCAN algorithm (#330, #332)
  • regression: add perceptron, PA, PA2, CW, AROW, NHERD, NN, cosine and euclidean (#327, #328)

• Improvements

  • clustering: improve APIs to make each Datum distinguishable by user-specified ID (#317, #326)
  • clustering: improve config format and parameter range checking (#314, #316, #325, #334, #335)
  • graph: add method to return the largest ID (#333)
  • classifier and regression: Rename diff_size to num_features_diff status key (#336)
  • Improve binary feature key names to include the rule name (#329)
  • regression: remove unused argument from driver constructor (#319, #320)
  • Remove unused revert function (#315, #331)

• Bug fixes

  • recommender: fix rows not cleared occasionally in inverted_index and inverted_index_euclid (#318)

Release 0.3.4 - 2016/08/29

• Improvements

  • recommender: add ignore_orthogonal option to inverted_index_euclid (#305, #309)
  • Exception error messages are now displayed in clean format (#313)
  • Remove Glibc C++ debug flag in debug mode (#306)

• Bug fixes

  • classifier: fix deadlock in nearest_neighbor-based classifier (#303)
  • Fix out-of-bound index access while MIXing column table (#307, #310)

Release 0.3.3 - 2016/07/25

• Improvements

  • Support UTF-8 strings in char_splitter (#299, #304)
  • Implement IDF-1 global weighting to fv_converter (#209, #300)

• Bug fixes

  • nearest_neighbor: fix deadlock issue (#302)

Release 0.3.2 - 2016/06/27

• Improvements

  • Implement Okapi BM25 global weighting to fv_converter (#284, #295)
  • Implement split string feature extractor to fv_converter (#255, #294)
  • Update README.rst (#297)

• Bug fixes

  • clustering: fix clear RPC to reset revision number (#164, #296)

Release 0.3.1 - 2016/05/30

• New feature

  • weight: driver that returns raw results of fv_converter (#288)

• Improvements

  • nearest_neighbor: improve performance by using LRU caching (#247)
  • nearest_neighbor: improve distance calculation precision in euclid_lsh (#253, #289)
  • recommender: support multi-threaded nearest neighbor search (#265)
  • recommender: improve distance calculation precision in inverted_index and inverted_index_euclid (#251, #290)
  • anomaly: support unlearning in lof algorithm (#231, #292)
  • Improve performance of LRU library (#266)
  • Improve compatibility with FreeBSD 10 (#280)

• Bug fixes

  • nearest_neighbor: fix thread safetiness issue in euclid_lsh (#283, #285)

Release 0.3.0 - 2016/04/25

• 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)

Release 0.2.9 - 2016/03/28

• Improvements

  • Expose fv_converter APIs (#256)

• Bug fixes

  • Fix bit_vector memory allocation size (#240)
  • Fix combination_rules to support except_{left,right} (#232, #258)

Release 0.2.8 - 2016/02/29

• Improvements

  • recommender: support unlearning in inverted_index_euclid algorithm (#237)

• Bug fixes

  • classifier: fix updating the same weight column for two times (#233, #236)
  • Code cleanup (#229)

Release 0.2.7 - 2016/01/25

• Improvements

  • recommender: add new recommender method (inverted_index_euclid) (#127)
  • recommender: support unlearning in MinHash algorithm (#220)
  • anomaly: support ignore_kth_same_option in LOF algorithm (#222, #223)

• Bug fixes

  • classifier: NN-based classifier does not take lock on label manipulation (#213, #221)
  • Fix key manager to swap next_id (#224, #226)

Release 0.2.6 - 2015/12/25

• Improvements

  • bandit: improve performance of select_arm API in ucb1 (#186, #214)

• Bug fixes

  • recommender: fix bit_index_storage (used by lsh/minhash) to ignore 0-bit vectors when analyzing (#211, #215)
  • fix typo in exception messages (#212)

Release 0.2.5 - 2015/11/30

• Improvements

  • recommender: support unlearning in LSH algorithm (#190)
  • recommender: fix typo of calc_similarity function name (#206, #208)
  • clustering: return error on calling get_k_center / get_core_members when clustering is not yet performed (#185, #207)
  • graph: improve error message (#203)
  • code cleanup (#205)

• Bug fixes

  • anomaly: fix crash when ignore_kth_same_point is not specified (#201)
  • nearest_neighbor: fix deadlock when running update/analyze simultaneously (#199, #200)

Release 0.2.4 - 2015/10/26

• Improvements

  • nearest_neighbor: improve performance of bit_vector based methods (#188, #191, #192, #193, #194, #196)
  • nearest_neighbor: improve get_all_row_ids performance by eliminating unnecessary lock (#188, #196)
  • classifier: improve performance by using read/write lock instead of exclusive lock (#184, #197)

• Bug fixes

  • nearest_neighbor: fix bit_vector bounds checking (#198)
  • nearest_neighbor: fix bit_vector to use GCC built-in popcount only when performance improvements are expected (#188, #196)
  • nearest_neighbor: fix missing lock in pack/unpack (#188, #196)

Release 0.2.3 - 2015/09/28

• Improvements

  • anomaly: add ignore_kth_same_point configuration option to avoid score to become inf (#130, #134)
  • clustering: add seed configuration option (#176, #180)
  • Improve unlearner overflow error message (#178, #187)
  • Code cleanup (#179)

• Bug fixes

  • classifier: fix missing lock in local_storage_mixture::inp (#182, #183)

Release 0.2.2 - 2015/08/31

• Improvements

  • clustering: improved so that consistent results are returned across runs (#167, #172)
  • clusteirng: add utility tests (#173)

• Bug fixes

  • clustering: fix test conditions (#129, #174)
  • clustering: fix MIX not working as expected (#70, #175)
  • bandit: fix gamma parameter validation (#163, #169)
  • fix bit_vector parameter assertion (#170)

Release 0.2.1 - 2015/07/27

• Improvements

  • recommender: support unlearner in inverted_index algorithm (#51, #120)
  • classifier: expose more status values (#166)
  • Support building with Python 3.x (#30, #162)
  • Code cleanup (#161)

• Bug fixes

  • bandit: fix exp3 algorithm calculations (#157, #158)
  • bandit: fix ucb1 algorithm calculations (#159, #160)

Release 0.2.0 - 2015/06/29

• Improvements

  • Support non-commutative functions in combination feature (#152, #156)
  • anomaly: lof algorithm now supports overwrite method (#154)
  • classifier: driver is now thread-safe (#144)
  • Improved compatibility with GCC-5 (#155)
  • Code cleanup (#140)

• Bug fixes

  • clustering: fix k-means segmentation fault when get_nearest_center is called before clustering is performed (#150, #151)
  • Column tables now touches unlearner on MIX (#100, #113)

Release 0.1.2 - 2015/04/27

• Improvements

  • bandit: add assume_unrewarded option (#125, #133)
  • Improved performance of bit_vector calculations (#137)
  • Codes cleanup (#132, #141, #143)

• Bug fixes

  • bandit: reject specifying unknown arm ID in register_reward API (#138, #148)
  • bandit: fix clear API to reset arm IDs (#142, #149)
  • recommender: fix retain_projection option not working in euclid_lsh (#98, #116)
  • Fix fv_converter become unresponsive when empty datum is given (#146, #147)
  • plugin.hpp is missing from installation (#139, #145)

Release 0.1.1 - 2015/03/30

• Improvements

  • Move column storage directory (#118, #123)
  • classifier: install algorithm headers (#117)
  • fv_converter reports error precisely (#119)
  • Codes / comments cleanup (#96, #97, #107, #109, #114, #115, #122, #135, #136)
  • Add language declarations to wscript (#108)
  • Update to the latest waf-unittest (#128)

• Bug fixes

  • clustering: avoid clusteirng score to become NaN (#78)

Release 0.1.0 - 2015/02/23

• New machine learning service

  • Distributed Multi-Armed Bandit (#111)

• Improvements

  • Add combination feature (#104)
  • classifier: Add NN-based method (#83)
  • clustering: Add test for gmm-based clustering algorithm (#66)
  • nearest_neighbor: implement get_all_rows API (#58, #101)
  • Update copyright (#103, #105)

• Bug fixes

  • weight_manager now MIX correctly in recommender, nearest_neighbor, anomaly and clustering (#61, #64)
  • weight_manager is now saved in nearest_neighbor and clustering (#62, #64)
  • nearest_neighbor: weight_manager is now updated when calling set_row API (#99, #64)
  • nearest_neighbor: overwriting rows now correctly updates the specified row (#84)
  • anomaly: neighbor of updated ID is not touched when using unlearner (#92, #94)

Release 0.0.7 - 2014/12/22

• Improvements

  • Remove error-prone default constructor (#72, #80)
  • clustering: Add test for gmm-based clustering algorithm (#66)

• Bug fixes

  • random_unlearner now take care of entries deleted by user (#60, #79)
  • Fix error message in gaussian_normalization_filter (#85, #86)

Release 0.0.6 - 2014/11/25

• Improvements

  • Add normalization feature (num_filter) to fv_converter (#67, #68)

  • recommender: Improve performance of inverted_index (#44, #45)

    • This work was supported by New Energy and Industrial Technology Development Organization (NEDO).

  • clustering: Support clear RPC method (#69)
  • burst: Improved parameter validation (#75, #77)
  • burst: Remove unused debug code (#71, #74)
  • Enable libstdc++ debug mode when configured using --enable-debug (#73)

• Bug fixes

  • recommender: Fix unlearner leaks rows when using NN-based method (fix #76)

Release 0.0.5 - 2014/10/20

• New machine learning service

  • Distributed Burst Detection

• Improvements

  • Removed an unnecessary typedef (#37)

Release 0.0.4 - 2014/09/29

• Improvements

  • Support string replacement with capture group in oniguruma-based regexp string_filter (#53)
  • Improve varidation of replacement string in re2-based regexp string_filter (#54)
  • classifier: Improve error message when invalid configuration is given (#52)

Release 0.0.3 - 2014/08/25

• Improvements

  • Improved speed of clustering test (#48)
  • anomaly: Exposed is_updatable interface in driver (#41)

• Bug fixes

  • classifier: Fix NHERD equations (#47)
  • stat: Fix wrong error message (#42)

Release 0.0.2 - 2014/07/22

• Improvements

  • Support for OS X Mavericks (#20, #23)
  • Make parameter name consistent with config (#38)
  • Add ABI version number to object file (#39)
  • #34, #35, #36, #40

• Bug fixes

  • classifier, regression: Fix a misused iterator (#21)
  • Remove out-of-bound access to vectors (#25)
  • Make bit_vector safer (#33)
  • Fix test for x86 machine (#31)

Release 0.0.1 - 2014/06/23

• First release of jubatus_core; algorithm component of Jubatus.

• jubatus_core was separated from Jubatus 0.5.4. Changes since 0.5.4 are as follows:

  • Add unlearning feature with sticky ID handling (#4)
  • Users can now disable regexp libraries at compile time (#8)
  • Renamed methods in Nearest Neighbor module: {similar,neighbor}_row_from_data is now called {similar,neighbor}_row_from_datum (#6)
  • Message improvements (#12)
  • Support for OS X Mavericks (#11)