Py-CompKey-Alg/ana_eval_sparse.py at master · wangprocess/Py-CompKey-Alg · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
# We reuse a fraction of code in http://bitbucket.org/omerlevy/hyperwords.
# Using the numpy and similarity matrix largely speed up the evaluation process,
# compared with evaluation scripts in word2vec and GloVe

import numpy as np
import argparse
import random
from scipy.sparse import dok_matrix, csr_matrix


def load_matrix(f_path):
    with open(f_path, errors='ignore') as f:
        row, col, data, iw = [], [], [], []
        first_line = True
        lines_num = 0
        for line in f:
            if first_line:
                first_line = False
                words_num = int(line.rstrip().split()[0])
                dim = int(line.rstrip().split()[1])
                continue
            line = line.rstrip().split(' ')
            word = line[0]
            iw.append(word)
            vector = line[1:]
            for v in vector:
                row.append(lines_num)
                col.append(int(v.split(":")[0]))
                data.append(float(v.split(":")[1]))
            lines_num += 1
        wi = {}
        for i in range(len(iw)):
            wi[iw[i]] = i
        row = np.array(row)
        col = np.array(col)
        data = np.array(data)
        matrix = csr_matrix((data, (row, col)), shape=(words_num, dim))
        return matrix, iw, wi


def load_vocabulary(path):
    with open(path) as f:
        vocab = [line.strip().split()[0] for line in f if len(line) > 0]
    return dict([(a, i) for i, a in enumerate(vocab)]), vocab


def read_analogy(path, iw):
    analogy = {}
    analogy_type = ""
    with open(path) as f:
        for line in f:
            oov = 0
            if line.strip().split()[0] == ':':
                analogy_type = line.strip().split()[1]
                analogy[analogy_type] = {}
                analogy[analogy_type]["questions"] = []
                analogy[analogy_type]["total"] = 0
                analogy[analogy_type]["seen"] = 0
                continue
            analogy_question = line.strip().split()
            for w in analogy_question[:3]:
                if w not in iw:
                    oov = 1
            if oov == 1:
                analogy[analogy_type]["total"] += 1
                continue
            analogy[analogy_type]["total"] += 1
            analogy[analogy_type]["seen"] += 1
            analogy[analogy_type]["questions"].append(analogy_question)

        for t in analogy:
            analogy[t]['iw'] = []
            analogy[t]['wi'] = {}
            for question in analogy[t]["questions"]:
                for w in question:
                    if w not in analogy[t]['iw']:
                        analogy[t]['iw'].append(w)
            for i, w in enumerate(analogy[t]['iw']):
                analogy[t]['wi'][w] = i
        return analogy


def normalize(matrix):
    matrix2 = matrix.copy()
    matrix2.data **= 2
    norm = np.reciprocal(np.sqrt(np.array(matrix2.sum(axis=1))[:, 0]))
    normalizer = dok_matrix((len(norm), len(norm)))
    normalizer.setdiag(norm)
    matrix = normalizer.tocsr().dot(matrix)
    return matrix


def guess(sims, analogy, analogy_type, iw, wi, word_a, word_b, word_c):
    sim_a = sims[analogy[analogy_type]["wi"][word_a]]
    sim_b = sims[analogy[analogy_type]["wi"][word_b]]
    sim_c = sims[analogy[analogy_type]["wi"][word_c]]

    add_sim = -sim_a+sim_b+sim_c
    add_sim[wi[word_a]] = 0
    add_sim[wi[word_b]] = 0
    add_sim[wi[word_c]] = 0
    guess_add = iw[np.nanargmax(add_sim)]

    mul_sim = sim_b * sim_c * np.reciprocal(sim_a+0.01)
    mul_sim[wi[word_a]] = 0
    mul_sim[wi[word_b]] = 0
    mul_sim[wi[word_c]] = 0
    guess_mul = iw[np.nanargmax(mul_sim)]
    return guess_add, guess_mul


def main():
    neg = 1
    vectors_path = "embedding_sample/sparse_small.txt"
    analogy_path = "CA8/morphological.txt"
    results = {}

    myParser = argparse.ArgumentParser()
    myParser.add_argument('-v', '--vectors', type=str, help="Vectors path")
    myParser.add_argument('-a', '--analogy', type=str, help="Analogy benchmark path")
    args = myParser.parse_args()
    if args.vectors:
        vectors_path = args.vectors
    if args.analogy:
        analogy_path = args.analogy

    matrix, iw, wi = load_matrix(vectors_path)  # Read matrix into the memory
    matrix = normalize(matrix)
    analogy = read_analogy(analogy_path, iw)
    for analogy_type in analogy.keys():  # Calculate the accuracy for each relation type
        correct_add_num, correct_mul_num = 0, 0
        analogy_matrix = matrix[[wi[w] if w in wi else random.randint(0, len(wi)-1) for w in analogy[analogy_type]["iw"]]]
        sims = analogy_matrix.dot(matrix.T)
        sims = np.array(sims.todense())
        for question in analogy[analogy_type]["questions"]:  # Loop for each analogy question
            word_a, word_b, word_c, word_d = question
            guess_add, guess_mul = guess(sims, analogy, analogy_type, iw, wi, word_a, word_b, word_c)
            if guess_add == word_d:
                correct_add_num += 1
            if guess_mul == word_d:
                correct_mul_num += 1
        cov = float(analogy[analogy_type]["seen"]) / analogy[analogy_type]["total"]
        if analogy[analogy_type]["seen"] == 0:
            acc_add = 0
            acc_mul = 0
            print (analogy_type + " add/mul: " + str(round(0.0, 3)) + "/" + str(round(0.0, 3)))
        else:
            acc_add = float(correct_add_num) / analogy[analogy_type]["seen"]
            acc_mul = float(correct_mul_num) / analogy[analogy_type]["seen"]
            print (analogy_type + " add/mul: " + str(round(acc_add, 3)) + "/" + str(round(acc_mul, 3)))
        # Store the results
        results[analogy_type] = {}
        results[analogy_type]["coverage"] = [cov, analogy[analogy_type]["seen"], analogy[analogy_type]["total"]]
        results[analogy_type]["accuracy_add"] = [acc_add, correct_add_num, analogy[analogy_type]["seen"]]
        results[analogy_type]["accuracy_mul"] = [acc_mul, correct_mul_num, analogy[analogy_type]["seen"]]

    correct_add_num, correct_mul_num, seen = 0, 0, 0
    for analogy_type in results:
        correct_add_num += results[analogy_type]["accuracy_add"][1]
        correct_mul_num += results[analogy_type]["accuracy_mul"][1]
        seen += results[analogy_type]["coverage"][1]

    # print results
    print("Total accuracy (add): " + str(round(float(correct_add_num)/seen, 3)))
    print("Total accuracy (mul): " + str(round(float(correct_mul_num)/seen, 3)))


if __name__ == '__main__':
    main()