Merge branch 'master' of ssh://github.com/honnibal/spaCy

fabfile.py

@@ -61,11 +61,14 @@ def docs():
         with lcd(path.join(path.dirname(__file__), 'docs')):
             local('make html')
 
-def publish():
+def publish(version):
     with virtualenv(VENV_DIR):
-        local('python setup.py register')
-        local('twine upload dist/*.tar.gz')
         local('git push origin master')
+        local('git tag -a %s' % version)
+        local('git push origin %s' % version)
+        local('python setup.py sdist')
+        local('python setup.py register')
+        local('twine upload dist/%s.tar.gz' % version)
 
 
 def env(lang="python2.7"):

setup.py

@@ -134,7 +134,7 @@ def run_setup(exts):
     headers_workaround.install_headers('numpy')
 
 
-VERSION = '0.91'
+VERSION = '0.93'
 def main(modules, is_pypy):
     language = "cpp"
     includes = ['.', path.join(sys.prefix, 'include')]

spacy/lexeme.pyx

@@ -43,11 +43,18 @@ cdef class Lexeme:
         return True if Lexeme.c_check_flag(self.c, flag_id) else False
 
     def similarity(self, other):
+        if self.vector_norm == 0 or other.vector_norm == 0:
+            return 0.0
         return numpy.dot(self.vector, other.vector) / (self.vector_norm * other.vector_norm)
 
     property has_vector:
         def __get__(self):
-            return sum(abs(self.c.repvec)) != 0
+            cdef int i
+            for i in range(self.vocab.vectors_length):
+                if self.c.repvec[i] != 0:
+                    return True
+            else:
+                return False
 
     property vector_norm:
         def __get__(self):

spacy/tokens/doc.pyx

@@ -127,6 +127,8 @@ cdef class Doc:
         return u''.join([t.string for t in self])
 
     def similarity(self, other):
+        if self.vector_norm == 0 or other.vector_norm == 0:
+            return 0.0
         return numpy.dot(self.vector, other.vector) / (self.vector_norm * other.vector_norm)
 
     property repvec:

spacy/tokens/spans.pyx

@@ -60,6 +60,8 @@ cdef class Span:
         self._seq.merge(self[0].idx, self[-1].idx + len(self[-1]), tag, lemma, ent_type)
 
     def similarity(self, other):
+        if self.vector_norm == 0.0 or other.vector_norm == 0.0:
+            return 0.0
         return numpy.dot(self.vector, other.vector) / (self.vector_norm * other.vector_norm)
 
     property vector:

spacy/tokens/token.pyx

@@ -50,6 +50,8 @@ cdef class Token:
         return self.doc[self.i+i]
 
     def similarity(self, other):
+        if self.vector_norm == 0 or other.vector_norm == 0:
+            return 0.0
         return numpy.dot(self.vector, other.vector) / (self.vector_norm * other.vector_norm)
 
     property lex_id:
@@ -130,7 +132,12 @@ cdef class Token:
 
     property has_vector:
         def __get__(self):
-            return sum(abs(self.c.lex.repvec)) != 0
+            cdef int i
+            for i in range(self.vocab.vectors_length):
+                if self.c.lex.repvec[i] != 0:
+                    return True
+            else:
+                return False
 
     property vector:
         def __get__(self):

spacy/vocab.pyx

@@ -1,6 +1,5 @@
 from __future__ import unicode_literals
 
-
 from libc.stdio cimport fopen, fclose, fread, fwrite, FILE
 from libc.string cimport memset
 from libc.stdint cimport int32_t
@@ -117,16 +116,14 @@ cdef class Vocab:
 
     cdef const LexemeC* _new_lexeme(self, Pool mem, unicode string) except NULL:
         cdef hash_t key
-        #cdef bint is_oov = mem is not self.mem
+        cdef bint is_oov = mem is not self.mem
-        # TODO
-        is_oov = False
-        mem = self.mem
         if len(string) < 3:
             mem = self.mem
         lex = <LexemeC*>mem.alloc(sizeof(LexemeC), 1)
         lex.orth = self.strings[string]
         lex.length = len(string)
         lex.id = self.length
+        lex.repvec = <float*>mem.alloc(self.vectors_length, sizeof(float))
         if self.get_lex_attr is not None:
             for attr, func in self.get_lex_attr.items():
                 value = func(string)
@@ -283,7 +280,7 @@ cdef class Vocab:
                                                            vec_len, len(pieces))
                 orth = self.strings[word_str]
                 lexeme = <LexemeC*><void*>self.get_by_orth(self.mem, orth)
-                lexeme.repvec = <float*>self.mem.alloc(len(pieces), sizeof(float))
+                lexeme.repvec = <float*>self.mem.alloc(self.vectors_length, sizeof(float))
 
                 for i, val_str in enumerate(pieces):
                     lexeme.repvec[i] = float(val_str)

tests/tokens/test_token_api.py

@@ -51,3 +51,14 @@ def test_is_properties(EN):
     assert addr.is_oov
     assert not Hi.is_oov
 
+@pytest.mark.models
+def test_vectors(EN):
+    apples, oranges, oov = EN(u'apples oranges ldskbjlsdkbflzdfbl')
+    assert apples.has_vector
+    assert oranges.has_vector
+    assert not oov.has_vector
+    assert apples.similarity(oranges) > apples.similarity(oov)
+    assert apples.similarity(oranges) == oranges.similarity(apples)
+    assert sum(apples.vector) != sum(oranges.vector)
+    assert apples.vector_norm != oranges.vector_norm
+    

tests/tokens/test_vec.py

@@ -4,15 +4,15 @@ from spacy.en import English
 
 import pytest
 
-@pytest.mark.vectors
+@pytest.mark.models
 def test_vec(EN):
     hype = EN.vocab['hype']
     assert hype.orth_ == 'hype'
-    assert 0.08 >= hype.repvec[0] > 0.07
+    assert 0.08 >= hype.vector[0] > 0.07
 
 
-@pytest.mark.vectors
+@pytest.mark.models
 def test_capitalized(EN):
     hype = EN.vocab['Hype']
     assert hype.orth_ == 'Hype'
-    assert 0.08 >= hype.repvec[0] > 0.07
+    assert 0.08 >= hype.vector[0] > 0.07