File size: 7,248 Bytes
e89a301 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 | import math
import numpy as np
import pandas as pd
from utils import rank, read_file, score, polartoscalar
from collections import Counter, defaultdict
# from scipy.special import comb
import timeit
from bisect import bisect
from copy import deepcopy
def ranking_sampled_vector(path, n_vectors, columns):
Theta=list(np.random.uniform(low=0,high=math.pi / 2, size=n_vectors))
Theta.insert(0,0)
dataset=read_file(path,columns)
R=[]
for theta in Theta:
r=deepcopy(rank(dataset,[theta],2))
R.append(r)
return R, Theta
def find_fair_ranking(path, n_vectors, columns, sens_attr_col, number_of_buckets):
start = timeit.default_timer()
R,Theta = ranking_sampled_vector(path, n_vectors, columns)
df = pd.read_csv(path)
G_=list(df[sens_attr_col].values)
G = [G_[i] for i in R[0]]
sens_attr_values = np.unique(df[sens_attr_col])
freq = Counter(G)
minority = min(freq, key=freq.get)
# distributions = []
collision_prob_dict = defaultdict(list)
for r in R:
bucket_size = len(r) // number_of_buckets
bucket_distribution = []
# collision_count = defaultdict(int)
for j in range(number_of_buckets):
bucket = []
for k in range(bucket_size):
bucket.append(G[r[j * bucket_size + k]])
bucket_distribution.append(bucket)
# bucket_distribution.append(
# [
# bucket.count(sens_attr)/bucket_size
# for sens_attr in sens_attr_values
# ]
# )
# for val in sens_attr_values:
# collision_count[val] += comb(
# bucket.count(val), len(sens_attr_values)
# )
# print(bucket)
for val in sens_attr_values:
collision_prob=0
for bucket in bucket_distribution:
# collision_prob_dict[val].append(
# collision_count[val] / comb(G.count(val), len(sens_attr_values))
# )
collision_prob+= (bucket.count(val)/freq[val])**2
collision_prob_dict[val].append(collision_prob)
# distributions.append(bucket_distribution)
disparity = []
for i in range(len(collision_prob_dict[minority])):
max_collision_prob_dict = np.max(
[collision_prob_dict[val][i] for val in sens_attr_values]
)
min_collision_prob_dict = np.min(
[collision_prob_dict[val][i] for val in sens_attr_values]
)
disparity.append((max_collision_prob_dict / min_collision_prob_dict) - 1)
stop = timeit.default_timer()
max_collision_prob_dict_original = np.max(
[collision_prob_dict[sens_attr][0] for sens_attr in sens_attr_values]
)
min_collision_prob_dict_original = np.min(
[collision_prob_dict[sens_attr][0] for sens_attr in sens_attr_values]
)
disparity_original = (max_collision_prob_dict_original / min_collision_prob_dict_original) - 1
return (
min(disparity),
disparity_original,
R[disparity.index(min(disparity))],
Theta[disparity.index(min(disparity))],
stop - start,
)
def query(q, f, scores, d, number_of_buckets):
c = 0
for j in range(d):
c += f[j] * q[j]
hash_bucket = (bisect(scores, c) // (len(scores) // number_of_buckets)) + 1
return hash_bucket
def fit_predict_eval(path_train, path_test,n_vectors, column,sens_attr_col,num_of_buckets):
_, _, _, theta, _ = find_fair_ranking(path_train, n_vectors, column, sens_attr_col, num_of_buckets)
dataset = read_file(path_train, column)
f = polartoscalar([theta], 2)
scores = sorted([score(dataset[i], f, 2) for i in range(len(dataset))])
test=pd.read_csv(path_test)
dict=defaultdict(list)
G = list(test[sens_attr_col].values)
n = len(G)
freq = Counter(G)
sens_attr_values = np.unique(G)
for _, row in test.iterrows():
bucket = query([row[column[0]],row[column[1]]], f, scores, 2, num_of_buckets)
dict[bucket].append(row[sens_attr_col])
collision_prob_single = defaultdict(int)
collision_prob_pairwise = defaultdict(int)
collision_prob=0
for bucket in dict.values():
collision_prob += (len(bucket)/n)**2
for val in sens_attr_values:
collision_prob_single[val] += (bucket.count(val)/freq[val])*(len(bucket)/n)
collision_prob_pairwise[val] += (bucket.count(val)/freq[val])**2
max_collision_prob_single = np.max(
[collision_prob_single[sens_attr] for sens_attr in sens_attr_values]
)
min_collision_prob_single = np.min(
[collision_prob_single[sens_attr] for sens_attr in sens_attr_values]
)
single_fairness= (max_collision_prob_single / (min_collision_prob_single)) - 1
max_collision_prob_pairwise = np.max(
[collision_prob_pairwise[sens_attr] for sens_attr in sens_attr_values]
)
min_collision_prob_pairwise = np.min(
[collision_prob_pairwise[sens_attr] for sens_attr in sens_attr_values]
)
pairwise_fairness= (max_collision_prob_pairwise / (min_collision_prob_pairwise)) - 1
return collision_prob, single_fairness, pairwise_fairness
# return collision_prob, collision_prob_single, collision_prob_pairwise
def fit_predict_eval_input(path_train, path_test,n_vectors, column,sens_attr_col,num_of_buckets):
dataset = read_file(path_train, column)
f = polartoscalar([0], 2)
scores = sorted([score(dataset[i], f, 2) for i in range(len(dataset))])
test=pd.read_csv(path_test)
dict=defaultdict(list)
G = list(test[sens_attr_col].values)
n = len(G)
freq = Counter(G)
sens_attr_values = np.unique(G)
for _, row in test.iterrows():
bucket = query([row[column[0]],row[column[1]]], f, scores, 2, num_of_buckets)
dict[bucket].append(row[sens_attr_col])
collision_prob_single = defaultdict(int)
collision_prob_pairwise = defaultdict(int)
collision_prob=0
for bucket in dict.values():
collision_prob += (len(bucket)/n)**2
for val in sens_attr_values:
collision_prob_single[val] += (bucket.count(val)/freq[val])*(len(bucket)/n)
collision_prob_pairwise[val] += (bucket.count(val)/freq[val])**2
max_collision_prob_single = np.max(
[collision_prob_single[sens_attr] for sens_attr in sens_attr_values]
)
min_collision_prob_single = np.min(
[collision_prob_single[sens_attr] for sens_attr in sens_attr_values]
)
single_fairness= (max_collision_prob_single / (min_collision_prob_single)) - 1
max_collision_prob_pairwise = np.max(
[collision_prob_pairwise[sens_attr] for sens_attr in sens_attr_values]
)
min_collision_prob_pairwise = np.min(
[collision_prob_pairwise[sens_attr] for sens_attr in sens_attr_values]
)
pairwise_fairness= (max_collision_prob_pairwise / (min_collision_prob_pairwise)) - 1
return collision_prob, single_fairness, pairwise_fairness
# return collision_prob, collision_prob_single, collision_prob_pairwise
|