aaai24-code/utils.py at main · liamj2311/aaai24-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
from sklearn.metrics import recall_score
import numpy as np
import pandas as pd

# Recall below 25 percentile and above 75 percentile (top level vs rest of levels))
def compute_recall(data, categories, top_level):
    recall_dfs = []
    for category in categories:
        recall_results = {}
        for i in [1, 4]:
            condition_top = (data['score_MATq'] == i) & (data[category] == top_level)
            condition_low = (data['score_MATq'] == i) & (data[category] != top_level)
            scores_MATq_top = data.loc[condition_top, 'score_MATq']
            scores_MATq_low = data.loc[condition_low, 'score_MATq']

            for variable in ["pred1", "pred2", "pred3", "pred_C", "pred_X"]:
                scores_MAT_variable_top = data.loc[condition_top, f'score_MAT_{variable}']
                scores_MAT_variable_low = data.loc[condition_low, f'score_MAT_{variable}']

                recall_top = recall_score(scores_MATq_top, scores_MAT_variable_top, average='micro')
                recall_low = recall_score(scores_MATq_low, scores_MAT_variable_low, average='micro')

                recall_results[f'recall_top_{variable}_{i}'] = recall_top
                recall_results[f'recall_low_{variable}_{i}'] = recall_low

        recall_df = pd.DataFrame(list(recall_results.items()), columns=['Metric', 'Recall'])
        recall_df[['Group', 'Model', 'Percentile']] = recall_df['Metric'].str.extract(r'recall_(top|low)_(\w+)_(\d+)')
        recall_df['Recall'] = recall_df['Recall'].round(4)
        recall_df['Variable'] = category
        recall_df['Prediction'] = '25_75'
        recall_dfs.append(recall_df[['Group', 'Model', 'Percentile', 'Recall', 'Variable', 'Prediction']])
    return recall_dfs

# Recall between 25 percentile and 75 percentile (top level vs rest of levels))
def compute_recall_terciles(data, categories, top_level):
    recall_dfs = []
    for category in categories:
        recall_results = {}
        i = 2
        condition_top = (data['score_MATq'] == i) & (data[category] == top_level)
        condition_low = (data['score_MATq'] == i) & (data[category] != top_level)
        scores_MATq_top = data.loc[condition_top, 'score_MATq']
        scores_MATq_low = data.loc[condition_low, 'score_MATq']

        for variable in ["pred1_t", "pred2_t", "pred3_t", "pred_C_t", "pred_X_t"]:
            scores_MAT_variable_top = data.loc[condition_top, f'score_MAT_{variable}']
            scores_MAT_variable_low = data.loc[condition_low, f'score_MAT_{variable}']

            recall_top = recall_score(scores_MATq_top, scores_MAT_variable_top, average='micro')
            recall_low = recall_score(scores_MATq_low, scores_MAT_variable_low, average='micro')

            recall_results[f'recall_top_{variable}_{i}'] = recall_top
            recall_results[f'recall_low_{variable}_{i}'] = recall_low

        recall_df = pd.DataFrame(list(recall_results.items()), columns=['Metric', 'Recall'])
        recall_df[['Group', 'Model', 'Tercile']] = recall_df['Metric'].str.extract(r'recall_(top|low)_(\w+)_(\d+)')
        recall_df['Recall'] = recall_df['Recall'].round(4)
        recall_df['Variable'] = category
        recall_df['Prediction'] = 'between25_75'
        recall_dfs.append(recall_df[['Group', 'Model', 'Tercile', 'Recall', 'Variable', 'Prediction']])
    return recall_dfs

# Recall below and above the median (top level vs rest of levels))
def compute_recall_median(data, categories, top_level):
    recall_dfs = []
    score_pairs = [(1, 2), (3, 4)]
    for category in categories:
        recall_results = {}
        for pair in score_pairs:
            condition_top = ((data['score_MATq'] == pair[0]) | (data['score_MATq'] == pair[1])) & (data[category] == top_level)
            condition_low = ((data['score_MATq'] == pair[0]) | (data['score_MATq'] == pair[1])) & (data[category] != top_level)
            scores_MATq_top = data.loc[condition_top, 'score_MATq']
            scores_MATq_low = data.loc[condition_low, 'score_MATq']
            scores_MATq_top_binary = scores_MATq_top.apply(lambda x: 1 if x in pair else 0)
            scores_MATq_low_binary = scores_MATq_low.apply(lambda x: 1 if x in pair else 0)

            for variable in ["pred1", "pred2", "pred3", "pred_C", "pred_X"]:
                scores_MAT_variable_top = data.loc[condition_top, f'score_MAT_{variable}']
                scores_MAT_variable_low = data.loc[condition_low, f'score_MAT_{variable}']
                scores_MAT_variable_top_binary = scores_MAT_variable_top.apply(lambda x: 1 if x in pair else 0)
                scores_MAT_variable_low_binary = scores_MAT_variable_low.apply(lambda x: 1 if x in pair else 0)

                recall_top = recall_score(scores_MATq_top_binary, scores_MAT_variable_top_binary, average='binary')
                recall_low = recall_score(scores_MATq_low_binary, scores_MAT_variable_low_binary, average='binary')

                recall_results[f'recall_top_{variable}_{pair[0]}_{pair[1]}'] = recall_top
                recall_results[f'recall_low_{variable}_{pair[0]}_{pair[1]}'] = recall_low

        recall_df = pd.DataFrame(list(recall_results.items()), columns=['Metric', 'Recall'])
        recall_df[['Group', 'Model', 'Pair1', 'Pair2']] = recall_df['Metric'].str.extract(r'recall_(top|low)_(\w+)_(\d+)_(\d+)')
        recall_df['Recall'] = recall_df['Recall'].round(4)
        recall_df['Variable'] = category
        recall_df['Prediction'] = 'median'
        recall_dfs.append(recall_df[['Group', 'Model', 'Pair1', 'Pair2', 'Recall', 'Variable', 'Prediction']])
    return recall_dfs

# Calculate equalized odds for each variable
def calculate_odds(row_value, top_value):
    return row_value / top_value if top_value != 0 else None

#############

def mld(vals):
    sum = 0
    avg = np.mean(vals)
    for val in vals:
        if val == 0.0:
            sum += 0
        else:
            sum += np.log(avg/val)
    return sum/len(vals)

def gini(vals):
    # this function is based on the third equation in
    # https://www.statsdirect.com/help/default.htm#nonparametric_methods/gini.htm
    sorted_vals = sorted(vals, reverse=False)
    # values cannot be 0
    sorted_vals = [x + 0.0000001 for x in sorted_vals]
    # multiply by 100 each probability to have numerical stability
    # with values between 0 and 1 and n > 2 negative values were being produced
    sorted_vals = [x * 100 for x in sorted_vals]
    den = len(sorted_vals) * sum(sorted_vals)
    num = 0
    for i, v in enumerate(sorted_vals):
        num += (2*(i+1) - len(sorted_vals) - 1) * v
    return num/den

def expected_value(probs, labels):
    return np.sum(np.dot(np.array(probs), np.array(labels))) / len(probs)

def compute_moment(moment: str, probs, labels):
    if moment == "mean":
        return expected_value(probs=probs, labels=labels)

def iop(df: pd.DataFrame, sensitive_attr: str, labels = [0, 1], moment="mean", ineq_index="mld"):
    sensitive_vals = np.unique(df[sensitive_attr].values[~np.isnan(df[sensitive_attr].values)])
    val_to_count = {val: df.loc[df[sensitive_attr] == val].shape[0] for val in sensitive_vals}
    moments = []
    for val in sensitive_vals:
        probs = []
        for label in labels:
            probs.append(df.loc[(df[sensitive_attr] == val) & (df["label"] == label)].shape[0] / df.shape[0])
        w = val_to_count[val] / df.shape[0]
        moments.append(w * compute_moment(moment=moment, probs=probs, labels=labels))
    if ineq_index == "mld":
        return mld(moments)
    elif ineq_index == "gini":
        return gini(moments)
    else:
        return None

def binarise_predictions(preds: pd.Series, percentile_range: str):
    percentiles = pd.qcut(preds, 4, labels=[x for x in range(1, 5)])
    if percentile_range == "below-25":
        return pd.Series((percentiles <= 1).astype(int))
    elif percentile_range == "above-75":
        return pd.Series((percentiles > 3).astype(int))
    elif percentile_range == "between-25-75":
        res = pd.Series((percentiles > 1) & (percentiles < 4)).astype(int)
        return res

def print_iop(iop_val, percentile_range, sensitive_attr):
    pr_verbose = {
        "below-25": "below the 25th percentile",
        "above-75": "above the 75th percentile",
        "between-25-75": "between the 25th and the 75th percentile"
    }

    print(f"IOP value for predictions {pr_verbose[percentile_range]} for groups identified by values of sensitive attribute {sensitive_attr}: {iop_val}")