robust-poststratification/weights.py at master · wwbp/robust-poststratification · 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
180
181
from itertools import product

import numpy as np
import pandas as pd

import utils


def create_user_weights(user_data, bins, user_percents, population_percents, dem, smoothing=0, smooth_before_binning=False, uninformed_smoothing=False):
    total_twitter = user_data.shape[0]
    ii = len(bins) - 2
    if len(population_percents) == 1 and population_percents[0] == 1:
        user_data['weight'] = 1
        return user_data

    for bin_entry in bins[1:][::-1]:
        twitter_bin_n = round(user_percents[ii] * total_twitter, 0)
        if uninformed_smoothing:
            percentage_twitter_bin = (twitter_bin_n + 1) / (total_twitter + len(user_percents))
        elif smooth_before_binning:
            percentage_twitter_bin = twitter_bin_n / total_twitter
        else:
            percentage_twitter_bin = (twitter_bin_n + smoothing * population_percents[ii]) / (total_twitter + smoothing)

        if percentage_twitter_bin > 0:
            w = population_percents[ii] / percentage_twitter_bin
        else:
            w = 0

        if ii == len(bins) - 2:
            user_data['weight'] = np.where(user_data[dem] < bin_entry, w, None)
        else:
            user_data['weight'] = np.where(user_data[dem] < bin_entry, w, user_data['weight'])
        ii -= 1

    # fill missing entries and renormalize
    user_data['weight'].fillna(user_data['weight'].mean(), inplace=True)
    user_data['weight'] = user_data['weight'] / user_data['weight'].sum() * len(user_data)

    return user_data


def create_weights_single(user_data, population_data, dem, smoothing, min_bin_num, smooth_before_binning, uninformed_smoothing, user_bins, population_cols):
    bins, user_percents, population_percents = utils.get_bins(
        user_data=user_data,
        population_data=population_data,
        dem=dem,
        population_table_cols=population_cols,
        user_dem_bins=user_bins,
        smoothing=smoothing,
        min_bin_num=min_bin_num,
        smooth_before_binning=smooth_before_binning,
    )
    user_weights = create_user_weights(
        user_data=user_data,
        bins=bins,
        user_percents=user_percents,
        population_percents=population_percents,
        dem=dem,
        smoothing=smoothing,
        smooth_before_binning=smooth_before_binning,
        uninformed_smoothing=uninformed_smoothing,
    )

    return user_weights


def create_weights_rake(user_data, population_data, demographics, smoothing, min_bin_num, smooth_before_binning, uninformed_smoothing, user_dem_bins, population_dem_cols):
    dataset, user_dem_percents, population_dem_percents = utils.create_banded_dataset(
        user_data=user_data,
        population_data=population_data,
        demographics=demographics,
        smoothing=smoothing,
        min_bin_num=min_bin_num,
        smooth_before_binning=smooth_before_binning,
        user_dem_bins=user_dem_bins,
        population_dem_cols=population_dem_cols,
    )

    data = dataset.data()
    dataframe_data = {}
    columns = demographics + ['perc']

    bands = [dem + '_banded' for dem in demographics]
    group_dict = data.groupby(bands).agg(['count'])['user_id'].to_dict()['count']
    sorted_keys = list(product(*[population_dem_percents[dem].keys() for dem in demographics]))
    total_twitter_users = float(data.shape[0])
    naive_percentages = {key: np.prod([population_dem_percents[dem][k] for dem, k in zip(demographics, key)]) / (100 ** len(demographics)) for key in sorted_keys}

    for i, key in enumerate(sorted_keys):
        if key in group_dict:
            if uninformed_smoothing:
                num = group_dict[key] + 1
                den = float(total_twitter_users + len(naive_percentages))
            else:
                num = group_dict[key] + smoothing * naive_percentages[key]
                den = float(total_twitter_users + smoothing)

            dataframe_data[i] = list(key) + [num / den * 100]
        elif smoothing > 0 or uninformed_smoothing:
            if uninformed_smoothing:
                num = 1
                den = float(len(naive_percentages))
            else:
                num = smoothing * naive_percentages[key]
                den = float(smoothing)
            dataframe_data[i] = list(key) + [num / den * 100]
        else:
            dataframe_data[i] = list(key) + [0]

    rake_df = pd.DataFrame.from_dict(dataframe_data, orient='index')
    rake_df.columns = columns

    utils.rake(rake_df, population_dem_percents)
    rake_df.columns = bands + ['perc']
    user_weights = pd.merge(data, rake_df, on=bands)
    user_weights.rename(columns={'perc': 'weight'}, inplace=True)

    return user_weights


def create_weights_naive(user_data, population_data, demographics, smoothing, min_bin_num, smooth_before_binning, uninformed_smoothing, user_dem_bins, population_dem_cols):
    dataset, user_dem_percents, population_dem_percents = utils.create_banded_dataset(
        user_data=user_data,
        population_data=population_data,
        demographics=demographics,
        smoothing=smoothing,
        min_bin_num=min_bin_num,
        smooth_before_binning=smooth_before_binning,
        user_dem_bins=user_dem_bins,
        population_dem_cols=population_dem_cols,
    )

    data = dataset.data()
    data['naive_banded'] = 0
    combined_targets = {'naive_banded': {}}
    bands = [dem + '_banded' for dem in demographics]
    group_dict = data.groupby(bands).agg(['count'])['user_id'].to_dict()
    sorted_keys = list(product(*[population_dem_percents[dem].keys() for dem in demographics]))

    i = 0
    skipped = False
    for key in sorted_keys:
        count = group_dict['count'].get(key, 0)
        if count < 1:
            skipped = True
            continue

        prod = 1
        query = []
        for k, dem, band in zip(key, demographics, bands):
            prod *= population_dem_percents[dem][k]
            query.append(f'{band} == {k}')
        combined_targets['naive_banded'][i + 1] = prod
        data.loc[data.eval(' and '.join(query)), 'naive_banded'] = i + 1
        i += 1

    s = sum(combined_targets['naive_banded'].values())
    if skipped or round(s) != 100:
        # renormalize combined_targets
        combined_targets['naive_banded'] = {k: v * 100 / s for k, v in combined_targets['naive_banded'].items()}

    sorted_bins = sorted(list(combined_targets['naive_banded'].keys()))
    sorted_targets = [combined_targets['naive_banded'][kkey] for kkey in sorted_bins]
    sorted_sample_targets = data.groupby(['naive_banded']).agg(['count'])['user_id'].to_dict()['count']
    t = float(sum(sorted_sample_targets.values()))
    sorted_sample_targets = {k: v / t for k, v in sorted_sample_targets.items()}
    sorted_sample_targets = [sorted_sample_targets[key] for key in sorted_bins]

    user_weights = create_user_weights(
        user_data=data,
        bins=sorted_bins,
        user_percents=sorted_sample_targets,
        population_percents=sorted_targets,
        dem='naive_banded',
        smoothing=smoothing,
        smooth_before_binning=smooth_before_binning,
        uninformed_smoothing=uninformed_smoothing
    )

    return user_weights