[HIVEMALL-91] Implement Online LDA

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.5.0
• Assignee: Takuya Kitazawa
• Reporter: Makoto Yui
• Created: 17/Mar/17 05:02
• Updated: 18/Sep/18 05:26
• Resolved: 20/Apr/17 12:08

Description

Implement OnlineLDA [1,2].

Online Learning for Latent Dirichlet Allocation
[1] http://dl.acm.org/citation.cfm?id=2997285
https://wellecks.wordpress.com/2014/10/26/ldaoverflow-with-online-lda/
http://mlwave.com/tutorial-online-lda-with-vowpal-wabbit/
https://github.com/miberk/jolda
https://github.com/blei-lab/onlineldavb
http://scikit-learn.org/stable/modules/generated/sklearn.decomposition.LatentDirichletAllocation.html

Streaming LDA is improved version of online LDA.
https://github.com/jessykate/streamLDA
[2] http://kzhai.github.io/paper/2013_icml.pdf

Rush implementation
https://github.com/NaokiStones/hivemall/tree/dev/lda

Issue Links

• incorporates: HIVEMALL-120 Refactor on LDA/pLSA's mini-batch & buffered iteration logic
• is broken by: HIVEMALL-109 Fix multi-byte chars & encoding related issue on LDA/pLSA UDFs
• relates to: HIVEMALL-210 Bug in LDA/PLSA prediction logic
• relates to: HIVEMALL-219 NullPointerException in pLSA model
• relates to: HIVEMALL-218 Fixed train_lda NPE where input row is null
• links to: GitHub Pull Request #66