Optimize Native unsupervised FastText (#1742)

* adds fasttext extension to setup

* cythonizes training using skipgram with negative sampling

* loop over indexes using index iterator

* cythonizes training using skipgram with hierarchical softmax

* adds cython generated .c file

* resolves segmentation fault with multiple workers

* fixes accuracy issues due to reference counts of word_subwords becoming 0

* cythonizes fasttext cbow architecture

* cleans extra variables/values

* corrects parameters order for word_locks* in cbow

* fixes indentation, unused imports and logging warning for slow version

* splits long lines and removes redundant `import`/`else`

* minor: removes redundant `else`

* adds docstring

* changes docstrings style, splits long lines

* fix references in fasttext docstring

* adds deleted else in cbow-neg

* fixes docstring format

* adds missing docstrings

* add missing __getitem__ to rst

* add missing import to __init__ (`from gensim.models import FastText` instead of `from gensim.models.fasttext ...`)

* fix docs

MANIFEST.in

@@ -10,3 +10,5 @@ include gensim/models/word2vec_inner.pyx
 include gensim/models/word2vec_inner.pxd
 include gensim/models/doc2vec_inner.c
 include gensim/models/doc2vec_inner.pyx
+include gensim/models/fasttext_inner.c
+include gensim/models/fasttext_inner.pyx

docs/src/models/fasttext.rst

@@ -5,5 +5,6 @@
     :synopsis: FastText model
     :members:
     :inherited-members:
+    :special-members: __getitem__
     :undoc-members:
     :show-inheritance:

gensim/models/__init__.py

@@ -19,6 +19,7 @@
 from .normmodel import NormModel  # noqa:F401
 from .atmodel import AuthorTopicModel  # noqa:F401
 from .ldaseqmodel import LdaSeqModel  # noqa:F401
+from .fasttext import FastText  # noqa:F401
 
 from . import wrappers  # noqa:F401