s4-code/plot_frequency_shift_correlation.py at master · mgruppi/s4-code · 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
170
171
172
173
174
175
176
177
178
179
import pandas as pd
import os
import numpy as np
from scipy import stats
from scipy.stats import pearsonr
import re
import matplotlib.pyplot as plt
import matplotlib
import seaborn as sns
from direction_analysis import load_experiment_results
from frequency_norm_experiment import load_word_counts


if __name__ == "__main__":
    exp_path = 'results/direction/'
    files = sorted(os.listdir(exp_path))
    output_path = 'results/frequency-shift'

    if not os.path.exists(output_path):
        os.makedirs(output_path)

    # df = pd.read_csv(os.path.join(exp_path, files[0]))

    data = {'k': list(), 'overlap': list(), 'dataset': list()}

    chosen_datasets = {
        # 'bnc-coca': 'UKUS English',
        # 'english-corpus1-english-corpus2': "English(SE2020)",
        # 'german-corpus1-german-corpus2': 'German',
        # 'latin-corpus1-latin-corpus2': 'Latin',
        # 'swedish-corpus1-swedish-corpus2': 'Swedish',
        # 'spanish-old-spanish-modern': 'Spanish',
        # 'wacad1990-wfic1990': 'Academic-Fiction',
        # 'wacad2000-wspok2000': 'Academic-Spoken',
        # 'wacad2012-wnews2012': 'Academic-News',
        # 'wacad2012-wfic2012': 'Academic-Fiction',
        # 'cs.AI-physics.class-ph': 'Ai-Physics (ArXiV)'
        'wacad1990-wacad2012': 'Academic 1990-2012',
        'wfic1990-wfic2012': 'Fiction 1990-2012',
        'wmag1990-wmag2012': 'Magazines 1990-2012',
        'wnews1990-wnews2012': 'News 1990-2012'
    }

    cmap = matplotlib.cm.get_cmap('Set2')
    colors = [cmap(i) for i in range(20)]

    dataset_freqs = {
        'bnc-coca': {
            'a': 'results/word_counts/bnc.txt',
            'b': 'results/word_counts/coca.txt',
            'color': colors[0]
        },
        'english-corpus1-english-corpus2': {
            'a': 'results/word_counts/english-corpus1.txt',
            'b': 'results/word_counts/english-corpus2.txt',
            'color': colors[1]
        },
        'german-corpus1-german-corpus2': {
            'a': 'results/word_counts/german-old-dta.txt',
            'b': 'results/word_counts/german-modern-bznd.txt',
            'color': colors[2]
        },
        'latin-corpus1-latin-corpus2': {
            'a': 'results/word_counts/latin-LatinISE1.txt',
            'b': 'results/word_counts/latin-LatinISE2.txt',
            'color': colors[3]
        },
        'swedish-corpus1-swedish-corpus2': {
            'a': 'results/word_counts/swedish-old-kubhist2a.txt',
            'b': 'results/word_counts/swedish-modern-kubhist2b.txt',
            'color': colors[4]
        },
        'spanish-old-spanish-modern': {
            'a': 'results/word_counts/dataset_XIX_lemmatized.txt',
            'b': 'results/word_counts/modern_corpus_lemmatized.txt',
            'color': colors[5]
        },
        'wacad1990-wfic1990': {
            'a': 'results/word_counts/coca/w_acad_1990.txt',
            'b': 'results/word_counts/coca/w_fic_1990.txt',
            'color': colors[0]
        },
        'wacad2000-wspok2000': {
            'a': 'results/word_counts/coca/w_acad_2000.txt',
            'b': 'results/word_counts/coca/w_spok_2000.txt',
            'color': colors[1]
        },
        'wacad2012-wnews2012': {
            'a': 'results/word_counts/coca/w_acad_2012.txt',
            'b': 'results/word_counts/coca/w_news_2012.txt',
            'color': colors[2]
        },
        'wacad2012-wfic2012': {
            'a': 'results/word_counts/coca/w_acad_2012.txt',
            'b': 'results/word_counts/coca/w_fic_2012.txt',
            'color': colors[3]
        },
        'wacad1990-wacad2012': {
            'a': 'results/word_counts/coca/w_acad_1990.txt',
            'b': 'results/word_counts/coca/w_acad_2012.txt',
            'color': colors[0]
        },
        'wfic1990-wfic2012': {
            'a': 'results/word_counts/coca/w_fic_1990.txt',
            'b': 'results/word_counts/coca/w_fic_2012.txt',
            'color': colors[1]
        },
        'wmag1990-wmag2012': {
            'a': 'results/word_counts/coca/w_mag_1990.txt',
            'b': 'results/word_counts/coca/w_mag_2012.txt',
            'color': colors[2]
        },
        'wnews1990-wnews2012': {
            'a': 'results/word_counts/coca/w_news_1990.txt',
            'b': 'results/word_counts/coca/w_news_2012.txt',
            'color': colors[4]
        },
    }

    # hue_order = ['English(SE2020)', 'German', 'Latin', 'Swedish', 'UKUS English', 'Spanish']
    hue_order = None
    sns.set_style("whitegrid")
    plt.rcParams.update({'font.size': 14})

    for i, f in enumerate(files):
        df, source_a, source_b = load_experiment_results(os.path.join(exp_path, f))
        ds = '%s-%s' % (source_a, source_b)

        print(" -", ds)
        df['freq'] = pd.Series([np.nan] * len(df))
        if ds in chosen_datasets:
            freq_a = load_word_counts(dataset_freqs[ds]['a'])
            freq_b = load_word_counts(dataset_freqs[ds]['b'])
            total_a = sum([v for v in freq_a.values()])
            total_b = sum([v for v in freq_b.values()])

            for row in df.itertuples():
                if row.direction == 'a_to_b':
                    if row.word in freq_a and row.word in freq_b:
                        # f = freq_b[row.word]/total_b - freq_a[row.word]/total_a
                        f = freq_a[row.word]+freq_b[row.word]
                        df.at[row.Index, 'freq'] = f
                else:
                    if row.word in freq_a and row.word in freq_b:
                        # f = freq_a[row.word]/total_a - freq_b[row.word]/total_b
                        f = freq_b[row.word]+freq_a[row.word]
                        df.at[row.Index, 'freq'] = f

            df = df.dropna()

            d = df[(df['direction'] == 'a_to_b') & (df['freq'] > 100)]
            df_sample = d.sample(n=500)
            pr = pearsonr(d['distance'], np.log(d['freq']))
            sns.scatterplot(data=df_sample, x='distance', y='freq', color=dataset_freqs[ds]['color'])
            plt.yscale('log')
            plt.title('r=%.2f' % pr[0])
            plt.ylabel('Frequency')
            plt.xlabel('Semantic Shift (cosine dist.)')
            plt.savefig(os.path.join(output_path, '%s.pdf' % ds))
            plt.close()
            print("---"*10)

    #     ds = '%s-%s' % (source_a, source_b)
    #     print("---"*10)
    #     print(ds)
    #     x_k, x_kn, x_o = get_shift_rankings(df)

    #     if ds in chosen_datasets:
    #         data['k'].extend(x_k)
    #         data['overlap'].extend(x_o)
    #         data['dataset'].extend([chosen_datasets[ds]] * len(x_k))

    # plt.rcParams.update({'font.size': 14})
    # sns.set_style("whitegrid")
    # sns.lineplot(data=data, x='k', y='overlap', hue='dataset', hue_order=hue_order, palette='Accent')
    # plt.xlabel('k (depth)')
    # plt.ylabel('Overlap')
    # plt.savefig('results/direction-overlap.pdf')