New method to get the word vector representation

glove/glove.py

@@ -285,17 +285,7 @@ def most_similar(self, word, number=5):
         Run a similarity query, retrieving number
         most similar words.
         """
-
-        if self.word_vectors is None:
-            raise Exception('Model must be fit before querying')
-
-        if self.dictionary is None:
-            raise Exception('No word dictionary supplied')
-
-        try:
-            word_idx = self.dictionary[word]
-        except KeyError:
-            raise Exception('Word not in dictionary')
+        word_idx = self._index_by_word(word)
 
         return self._similarity_query(self.word_vectors[word_idx], number)[1:]
 
@@ -307,3 +297,24 @@ def most_similar_paragraph(self, paragraph, number=5, **kwargs):
         paragraph_vector = self.transform_paragraph(paragraph, **kwargs)
 
         return self._similarity_query(paragraph_vector, number)
+
+    def word_vector_by_word(self, word):
+        """
+        Given a word returns its embedding vector representation
+        """
+        word_idx = self._index_by_word(word)
+
+        return self.word_vectors[word_idx]
+
+    def _index_by_word(self, word):
+        if self.word_vectors is None:
+            raise Exception('Model must be fit before querying')
+
+        if self.dictionary is None:
+            raise Exception('No word dictionary supplied')
+
+        try:
+            return self.dictionary[word]
+        except KeyError:
+            raise Exception('Word not in dictionary')
+

readme.md

@@ -59,6 +59,13 @@ Out[19]:
  ('racing', 0.83157724991920212)]
 ```
 
+You can also get the word vector representation:
+
+```
+In [20]: glove.word_vector_by_word('car')
+array([0.34795333, 0.63220108, 0.06546937, ... 0.66068305, 0.91771246, 0.01173065])
+```
+
 ## Development
 Pull requests are welcome.
 

tests/test_glove.py

@@ -5,6 +5,7 @@
 
 from utils import generate_training_corpus
 
+import pytest
 
 def _reproduce_input_matrix(glove_model):
 
@@ -80,3 +81,50 @@ def test_fitting():
     repr_matrix = _reproduce_input_matrix(glove_model)
 
     assert ((repr_matrix - log_cooc_mat) ** 2).sum() < 1500.0
+
+def test_word_vector_by_word():
+    glove_model = Glove()
+    glove_model.word_vectors = [
+      [1, 2, 3],
+      [4, 5, 6]
+    ]
+    glove_model.dictionary = {
+      "first": 0,
+      "second": 1
+    }
+
+    result0 = glove_model.word_vector_by_word("first")
+
+    assert(result0 == [1, 2, 3])
+
+    result1 = glove_model.word_vector_by_word("second")
+
+    assert(result1 == [4, 5, 6])
+
+def test_word_vector_by_word_without_fitting():
+    glove_model = Glove()
+    glove_model.dictionary = {"word": 0}
+
+    with pytest.raises(Exception) as ex:
+        glove_model.word_vector_by_word("word")
+
+    assert(ex.value.message == "Model must be fit before querying")
+
+def test_word_vector_by_word_without_dictionary():
+    glove_model = Glove()
+    glove_model.word_vectors = [[1, 2, 3]]
+
+    with pytest.raises(Exception) as ex:
+        glove_model.word_vector_by_word("word")
+
+    assert(ex.value.message == "No word dictionary supplied")
+
+def test_word_vector_by_word_without_word():
+    glove_model = Glove()
+    glove_model.word_vectors = [[1, 2, 3]]
+    glove_model.dictionary = {"other": 0}
+
+    with pytest.raises(Exception) as ex:
+        glove_model.word_vector_by_word("word")
+
+    assert(ex.value.message == "Word not in dictionary")