88
99
1010class GloVeModel :
11+ """Class to interact with GloVe embeddings.
12+
13+ Attributes
14+ ----------
15+ binarized_vectors : dict[str, list[str]]
16+ Dict of word: binarized_vector pairs.
17+ vec_file : str
18+ Path to GloVe embeddings file.
19+ vectors : dict[str, np.ndarray]
20+ Dict of word: vector pairs.
21+ """
22+
1123 def __init__ (self , vec_file : str ):
1224 self .vec_file = vec_file
1325 self ._load_vectors_from_file (vec_file )
26+ self ._binarize_vectors ()
1427
1528 def __repr__ (self ) -> str :
1629 return f"GloVeModel(vec_file={ self .vec_file }
@@ -51,7 +64,7 @@ def _load_vectors_from_file(self, vec_file: str) -> None:
5164 """Load vectors from gzipped txt file in word2vec format.
5265
5366 The first line of the file contains the header which is the vocabulary size
54- (i.e. number of vectors) and the dimenisions of the vectors.
67+ (i.e. number of vectors) and the dimensions of the vectors.
5568
5669 All remaining rows contain the token followed by the numeric elements of the
5770 vector, separated by a space
@@ -61,7 +74,7 @@ def _load_vectors_from_file(self, vec_file: str) -> None:
6174 vec_file : str
6275 File to load vectors from.
6376 """
64- vectors = {}
77+ self . vectors = {}
6578 with as_file (files (__package__ ) / vec_file ) as p :
6679 with gzip .open (p , "rt" ) as f :
6780 # Read first line as header
@@ -73,6 +86,41 @@ def _load_vectors_from_file(self, vec_file: str) -> None:
7386 parts = line .rstrip ().split ()
7487 token = parts [0 ]
7588 vector = np .array ([float (v ) for v in parts [1 :]], dtype = np .float32 )
76- vectors [token ] = vector
89+ self .vectors [token ] = vector
90+
91+ def _binarize_vectors (self ):
92+ """Binarize word vectors by converting continuous values into discrete values.
93+
94+ For each word vector, calculate the average value of the positive elements and
95+ the negative elements. Replace each element of each word vector according to:
96+ if value < negative_average:
97+ "NEG"
98+ elif value > positive_average
99+ "POS"
100+ else
101+ "0"
77102
78- self .vectors = vectors
103+ The resulting word vectors are stored in the binarized_vectors attribute.
104+
105+ References
106+ ----------
107+ J. Guo, W. Che, H. Wang, and T. Liu, ‘Revisiting Embedding Features for Simple
108+ Semi-supervised Learning’, in Proceedings of the 2014 Conference on Empirical
109+ Methods in Natural Language Processing (EMNLP), Doha, Qatar: Association for
110+ Computational Linguistics, 2014, pp. 110–120. doi: 10.3115/v1/D14-1012.
111+ """
112+ self .binarized_vectors = {}
113+ for word , vec in self .vectors .items ():
114+ positive_avg = np .mean (vec [vec > 0 ])
115+ negative_avg = np .mean (vec [vec < 0 ])
116+
117+ binarised_vec = []
118+ for value in vec :
119+ if value < negative_avg :
120+ binarised_vec .append ("VNEG" )
121+ elif value > positive_avg :
122+ binarised_vec .append ("VPOS" )
123+ else :
124+ binarised_vec .append ("V0" )
125+
126+ self .binarized_vectors [word ] = binarised_vec
0 commit comments