LMMS/exp_bias.py at master · rdenaux/LMMS · 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
import argparse
import logging
from collections import defaultdict

import numpy as np
from nltk.corpus import wordnet as wn

import matplotlib.pyplot as plt
from matplotlib import rc
rc('text', usetex=True)
plt.switch_backend('agg')
plt.rcParams.update({'font.size': 18})

from vectorspace import SensesVSM


logging.basicConfig(level=logging.DEBUG,
                    format='%(asctime)s - %(levelname)s - %(message)s',
                    datefmt='%d-%b-%y %H:%M:%S')


def wn_synset2sensekeys(synset):
    sensekeys = []
    for lemma in synset.lemmas():
        sensekeys.append(lemma.key())
    return sensekeys


def get_synset_vec(senses_vsm, synset):
    synset_vecs = []
    for synset_sensekey in wn_synset2sensekeys(synset):
        if synset_sensekey in senses_vsm.labels_set:
            synset_vecs.append(senses_vsm.get_vec(synset_sensekey))

    if len(synset_vecs) > 0:
        return np.mean(synset_vecs, axis=0)
    else:
        return None


def get_synset_name_tex(synset):
    name, pos, ind = synset.name().split('.')
    # return '$%s_{\small{%d}}^{\small{%s}}$' % (name, int(ind), pos)
    return '$%s_{\small{%s}}^{\small{%d}}$' % (name, pos, int(ind))


if __name__ == '__main__':

    parser = argparse.ArgumentParser(description='Bias Discovery Experiment with LMMS.', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    parser.add_argument('-lmms1024', help='Path to LMMS 1024 vectors', required=True)
    parser.add_argument('-lmms2048', help='Path to LMMS 2048 vectors', required=True)
    parser.add_argument('-gen_pdf', help='Generate Latex Chart', action='store_true', required=False)
    parser.set_defaults(gen_pdf=False)
    args = parser.parse_args()

    relevant_synsets = []
    relevant_synsets.append(wn.synset('programmer.n.01'))
    relevant_synsets.append(wn.synset('doctor.n.01'))
    relevant_synsets.append(wn.synset('doctor.n.04'))
    relevant_synsets.append(wn.synset('counselor.n.01'))
    relevant_synsets.append(wn.synset('counselor.n.02'))
    relevant_synsets.append(wn.synset('florist.n.01'))
    relevant_synsets.append(wn.synset('teacher.n.01'))
    relevant_synsets.append(wn.synset('nurse.n.01'))
    relevant_synsets.append(wn.synset('receptionist.n.01'))

    logging.info('Loading SensesVSM for LMMS 1024 ...')
    senses_vsm_1pt = SensesVSM(args.lmms1024)

    logging.info('Loading SensesVSM for LMMS 2048 ...')
    senses_vsm_2pt = SensesVSM(args.lmms2048)

    vsms = [senses_vsm_1pt, senses_vsm_2pt]

    if args.gen_pdf:
        plt.clf()
        fig, ax = plt.subplots()

        vsm_labels = ['LMMS$_{1024}$', 'LMMS$_{2048}$']
        vsm_colors = ['deepskyblue', 'royalblue']
        vsm_patterns = ['/', 'x']

    logging.info('Processing Sense Vectors ...')
    for sense_vsm_idx, senses_vsm in enumerate(vsms):

        vec_man = get_synset_vec(senses_vsm, wn.synset('man.n.01'))
        vec_woman = get_synset_vec(senses_vsm, wn.synset('woman.n.01'))

        scored_synsets = []
        for synset in relevant_synsets:
            vec_synset = get_synset_vec(senses_vsm, synset)
            score = np.dot(vec_man, vec_synset) - np.dot(vec_woman, vec_synset)
            logging.info('bias(%s) = %f' % (synset.name(), score))
            scored_synsets.append((synset, score))
        scored_synsets = sorted(scored_synsets, key=lambda x: x[1], reverse=True)

        if args.gen_pdf:
            ax.barh(np.arange(len(relevant_synsets)),
                    [score for synset, score in scored_synsets],
                    label=vsm_labels[sense_vsm_idx],
                    align='center',
                    alpha=0.5,
                    color=vsm_colors[sense_vsm_idx],
                    hatch=vsm_patterns[sense_vsm_idx])

    if args.gen_pdf:
        ax.set_yticks(np.arange(len(relevant_synsets)))
        ax.set_yticklabels([get_synset_name_tex(synset) for synset, score in scored_synsets])
        ax.invert_yaxis()
        # ax.set_xlabel('Gender Bias')
        plt.legend(loc='upper left')

        logging.info('Saving Bar Chart PDF at misc/bias.pdf ...')
        plt.savefig('misc/bias.pdf', bbox_inches='tight', format='pdf', dpi=300)