All Versions
82
Latest Version
Avg Release Cycle
82 days
Latest Release
734 days ago
Changelog History
Page 5
Changelog History
Page 5
-
v0.12.4 Changes
January 29, 2016- π Better internal handling of job batching in word2vec (#535)
- up to 300% speed up when training on very short documents (~tweets)
- Word2vec CLI in line with original word2vec.c (Andrey Kutuzov, #538)
- Same default values. See diff https://github.com/akutuzov/gensim/commit/6456cbcd75e6f8720451766ba31cc046b4463ae2
- Standalone script with command line arguments matching those of original C tool.
- Usage: python -m gensim.scripts.word2vec_standalone -train data.txt -output trained_vec.txt -size 200 -window 2 -sample 1e-4
- Improved load_word2vec_format() performance (@svenkreiss, #555)
- Remove
init_sims()
call for performance improvements when normalized vectors are not needed. - Remove
norm_only
parameter (API change). Callinit_sims(replace=True)
after theload_word2vec_format()
call for the oldnorm_only=True
behavior.
- Remove
- Word2vec allows non-strict unicode error handling (ignore or replace) (Gordon Mohr, #466)
- Doc2Vec
model.docvecs[key]
now raises KeyError for unknown keys (Gordon Mohr, #520) - Fix
DocvecsArray.index_to_doctag
somost_similar()
returns string doctags (Gordon Mohr, #560) - On-demand loading of the
pattern
library in utils.lemmatize (Jan Zikes, #461)utils.HAS_PATTERN
flag moved toutils.has_pattern()
- Threadsafe Word2Vec/Doc2Vec finish-check to avoid hang/unending Word2Vec/Doc2Vec training (Gordon Mohr, #571)
- Tuned
TestWord2VecModel.test_cbow_hs()
against random failures (Gordon Mohr, #531) - 0οΈβ£ Prevent ZeroDivisionError when
default_timer()
indicate no elapsed time (Gordon Mohr, #518) - Forwards compatibility for NumPy > 1.10 (Matti Lyra, #494, #513)
- LdaModel and LdaMulticore produce a large number of DeprecationWarnings from .inference() because the term ids in each chunk returned from utils.grouper are floats. This behaviour has been changed so that the term IDs are now ints.
- utils.grouper returns a python list instead of a numpy array in .update() when LdaModel is called in non distributed mode
- in distributed mode .update() will still call utils.grouper with as_numpy=True to save memory
- LdaModel.update and LdaMulticore.update have a new keyword parameter chunks_as_numpy=True/False (defaults to False) that allows controlling this behaviour
- π Better internal handling of job batching in word2vec (#535)
-
v0.12.3 Changes
November 05, 2015- π Make show_topics return value consistent across models (Christopher Corley, #448)
- All models with the
show_topics
method should return a list of(topic_number, topic)
tuples, wheretopic
is a list of(word, probability)
tuples. - This is a breaking change that affects users of the
LsiModel
,LdaModel
, andLdaMulticore
that may be reliant on the old tuple layout of(probability, word)
.
- All models with the
- Mixed integer & string document-tags (keys to doc-vectors) will work (Gordon Mohr, #491)
- DocvecsArray's
index2doctag
list is renamed/reinterpreted asoffset2doctag
offset2doctag
entries map todoctag_syn0
indexes after last plain-int doctag (if any)- (If using only string doctags,
offset2doctag
may be interpreted same asindex2doctag
.)
- DocvecsArray's
- π New Tutorials on Dynamic Topic Modelling and Classification via Word2Vec (@arttii #471, @mataddy #500)
- Auto-learning for the eta parameter on the LdaModel (Christopher Corley, #479)
- π Python 3.5 support
- Speed improvements to keyword and summarisation methods (@erbas #441)
- OSX wheels (#504)
- π Win build (#492)
- π Make show_topics return value consistent across models (Christopher Corley, #448)
-
v0.12.2 Changes
September 19, 2015- tutorial on text summarization (Γlavur Mortensen, #436)
- more flexible vocabulary construction in word2vec & doc2vec (Philipp Dowling, #434)
- β added support for sliced TransformedCorpus objects, so that after applying (for instance) TfidfModel the returned corpus remains randomly indexable. (Matti Lyra, #425)
- π changed the LdaModel.save so that a custom
ignore
list can be passed in (Matti Lyra, #331) - β added support for NumPy style fancy indexing to corpus objects (Matti Lyra, #414)
- py3k fix in distributed LSI (spacecowboy, #433)
- π Windows fix for setup.py (#428)
- π fix compatibility for scipy 0.16.0 (#415)
-
v0.12.1 Changes
July 20, 2015- π improvements to testing, switch to Travis CI containers
- π support for loading old word2vec models (<=0.11.1) in 0.12+ (Gordon Mohr, #405)
- π various bug fixes to word2vec, doc2vec (Gordon Mohr, #393, #386, #404)
- π TextSummatization support for very short texts (Federico Barrios, #390)
- π support for word2vec[['word1', 'word2'...]] convenience API calls (Satish Palaniappan, #395)
- π MatrixSimilarity supports indexing generator corpora (single pass)
-
v0.12.0 Changes
July 06, 2015- π complete API, performance, memory overhaul of doc2vec (Gordon Mohr, #356, #373, #380, #384)
- fast infer_vector(); optional memory-mapped doc vectors; memory savings with int doc IDs
- 'dbow_words' for combined DBOW & word skip-gram training; new 'dm_concat' mode
- multithreading & negative-sampling optimizations (also benefitting word2vec)
- API NOTE: doc vectors must now be accessed/compared through model's 'docvecs' field (eg: "model.docvecs['my_ID']" or "model.docvecs.most_similar('my_ID')")
- https://github.com/piskvorky/gensim/blob/develop/docs/notebooks/doc2vec-IMDB.ipynb
- π new "text summarization" module (PR #324: Federico Lopez, Federico Barrios)
- π new matutils.argsort with partial sort
- performance speedups to all similarity queries (word2vec, Similarity classes...)
- word2vec can compute likelihood scores for classification (Mat Addy, #358)
- π word2vec supports "encoding" parameter when loading from C format, for non-utf8 models
- more memory-efficient word2vec training (#385)
- π fixes to Python3 compatibility (Pavel Kalaidin #330, S-Eugene #369)
- β¨ enhancements to save/load format (Liang Bo Wang #363, Gordon Mohr #356)
- pickle defaults to protocol=2 for better py3 compatibility
- π fixes and improvements to wiki parsing (Lukas Elmer #357, Excellent5 #333)
- π fix to phrases scoring (Ikuya Yamada, #353)
- speed up of phrases generation (Dave Challis, #349)
- π changes to multipass LDA training (Christopher Corley, #298)
- π various doc improvements and fixes (Matti Lyra #331, Hongjoo Lee #334)
- π fixes and improvements to LDA (Christopher Corley #323)
- π complete API, performance, memory overhaul of doc2vec (Gordon Mohr, #356, #373, #380, #384)
-
v0.11.0 Changes
April 10, 2015- β added "topic ranking" to sort topics by coherence in LdaModel (jtmcmc, #311)
- π new fast ShardedCorpus out-of-core corpus (Jan Hajic jr., #284)
- utils.smart_open now uses the smart_open package (#316)
- π new wrapper for LDA in Vowpal Wabbit (Dave Challis, #304)
- π improvements to the DtmModel wrapper (Yang Han, #272, #277)
- π move wrappers for external modeling programs into a submodule (Christopher Corley, #295)
- π allow transparent compression of NumPy files in save/load (Christopher Corley, #248)
- πΎ save/load methods now accept file handles, in addition to file names (macks22, #292)
- π fixes to LdaMulticore on Windows (Feng Mai, #305)
- π lots of small fixes & py3k compatibility improvements (Chyi-Kwei Yau, Daniel Nouri, Timothy Emerick, Juarez Bochi, Christopher Corley, Chirag Nagpal, Jan Hajic jr., FlΓ‘vio CodeΓ§o Coelho)
- π re-released as 0.11.1 and 0.11.1-1 because of a packaging bug
-
v0.10.3 Changes
November 17, 2014- β added streamed phrases = collocation detection (Miguel Cabrera, #258)
- β added param for multiple word2vec epochs (sebastienj, #243)
- β added doc2vec (=paragraph2vec = extension of word2vec) model (Timothy Emerick, #231)
- π initialize word2vec deterministically, for increased experiment reproducibility (KCzar, #240)
- all indexed corpora now allow full Python slicing syntax (Christopher Corley, #246)
- β‘οΈ update distributed code for new Pyro4 API and py3k (Michael Brooks, Marco Bonzanini, #255, #249)
- π fixes to six module version (Lars Buitinck, #259)
- π fixes to setup.py (Maxim Avanov and Christopher Corley, #260, #251)
- β‘οΈ ...and lots of minor fixes & updates all around
-
v0.10.2 Changes
September 18, 2014- π new parallelized, LdaMulticore implementation (Jan Zikes, #232)
- Dynamic Topic Models (DTM) wrapper (Arttii, #205)
- word2vec compiled from bundled C file at install time: no more pyximport (#233)
- π¨ standardize show_/print_topics in LdaMallet (Benjamin Bray, #223)
- β add new word2vec multiplicative objective (3CosMul) of Levy & Goldberg (Gordon Mohr, #224)
- preserve case in MALLET wrapper (mcburton, #222)
- π support for matrix-valued topic/word prior eta in LdaModel (mjwillson, #208)
- π py3k fix to SparseCorpus (Andreas Madsen, #234)
- π fix to LowCorpus when switching dictionaries (Christopher Corley, #237)
-
v0.10.1 Changes
July 22, 2014- word2vec: new n_similarity method for comparing two sets of words (François Scharffe, #219)
- π¨ make LDA print/show topics parameters consistent with LSI (Bram Vandekerckhove, #201)
- β add option for efficient word2vec subsampling (Gordon Mohr, #206)
- π fix length calculation for corpora on empty files (Christopher Corley, #209)
- π improve file cleanup of unit tests (Christopher Corley)
- β more unit tests
- unicode now stored everywhere in gensim internally; accepted input stays either utf8 or unicode
- π various fixes to the py3k ported code
- π allow any dict-like input in Dictionary.from_corpus (Andreas Madsen)
- error checking improvements to the MALLET wrapper
- ignore non-articles during wiki parsig
- utils.lemmatize now (optionally) ignores stopwords
-
v0.10.0 Changes
June 04, 2014- π full Python 3 support (targeting 3.3+, #196)
- all internal methods now expect & store unicode, instead of utf8
- π new optimized word2vec functionality: negative sampling, cbow (sebastien-j, #162)
- allow by-frequency sort in Dictionary.save_as_text (Renaud Richardet, #192)
- β add topic printing to HDP model (Tiepes, #190)
- π new gensim_addons package = optional install-time Cython compilations (BjΓΆrn Esser, #197)
- β added py3.3 and 3.4 to Travis CI tests
- π fix a cbow word2vec bug (Liang-Chi Hsieh)