RECODE-H / annotations /annotation_32 /fairHashmap /ranking_sampled_vector_test.py
mcy98's picture
Upload folder using huggingface_hub
e89a301 verified
Raw History Blame Contribute Delete
11.8 kB
import timeit
import numpy as np
import pandas as pd
from ranking_sampled_vector import find_fair_ranking, query
from utils import read_file, score, polartoscalar, plot, plot_2
from pathlib import Path
from necklace_split_binary import necklace_split
import pickle
queries = []
for i in range(1000):
query_x = np.random.randint(0, 1)
query_y = np.random.randint(0, 1)
queries.append([query_x, query_y])
ratios = [0.25, 0.5, 0.75, 1.0]
n_vectors=[10,50,100,500,1000]
# n_vectors=[1000,1000,1000,1000,1000]
fractions = [0.2, 0.4, 0.6, 0.8, 1.0]
num_of_buckets_list = [100, 200, 300, 400, 500, 600, 700, 800, 900, 1000]
datasets = ["adult", "compas_random_id", "diabetes", "popsim_binary"]
sensitive_attrs = ["sex", "Sex_Code_Text", "gender", "race"]
columns = [
["fnlwgt", "education-num"],
["ID", "RawScore"],
["encounter_id", "patient_nbr"],
["lon", "lat"],
]
d = 2
for idx in range(len(datasets)):
preprocessing_time_t = []
query_times_t = []
disparities_after_t = []
disparities_before_t = []
print("=================", datasets[idx], "=================")
for i in range(5):
preprocessing_time = []
query_times = []
disparities_after = []
disparities_before = []
for frac in fractions:
print("=================", "fraction:", frac, "=================")
path = (
"real_data/"
+ datasets[idx]
+ "/"
+ datasets[idx]
+ "_f_"
+ str(frac)
+ ".csv"
)
n = pd.read_csv(path).shape[0]
num_of_buckets = 100
(
disparity,
disparity_original,
ranking,
theta,
duration,
) = find_fair_ranking(path, n_vectors[idx], columns[idx], sensitive_attrs[idx], num_of_buckets)
# print("Disparity:", disparity)
# print("Original Disparity:", disparity_original)
disparities_after.append(disparity)
disparities_before.append(disparity_original)
preprocessing_time.append(duration)
query_time = []
dataset = read_file(path, columns[idx])
f = polartoscalar([theta], d)
scores = sorted([score(dataset[i], f, d) for i in range(len(dataset))])
for q in queries:
start = timeit.default_timer()
query(q, f, scores, d, num_of_buckets)
stop = timeit.default_timer()
query_time.append(stop - start)
query_times.append(np.mean(query_time))
print("Varying dataset size (prep time):", preprocessing_time)
print("Varying dataset size (query time):", query_times)
print("Disparity Before:",disparities_before)
print("Disparity After:",disparities_after)
Path("plots/ranking_sampled_vector/" + datasets[idx]).mkdir(parents=True, exist_ok=True)
plot_2(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_size_unfairness.png",
fractions,
disparities_before,
disparities_after,
fractions,
"Varying dataset size (Disparity Before/After)",
"Fraction(×"+str(n)+")",
"Disparity Before/After",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_size_prep_time.png",
fractions,
preprocessing_time,
fractions,
"Varying dataset size (prep time)",
"Fraction(×"+str(n)+")",
"Time (sec)",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_size_query_time.png",
fractions,
query_times,
fractions,
"Varying dataset size (query time)",
"Fraction(×"+str(n)+")",
"Time (sec)",
[5e-7, 15e-7],
)
preprocessing_time = []
query_times = []
disparities_after = []
disparities_before = []
for ratio in ratios:
print("=================", "ratio:", ratio, "=================")
path = (
"real_data/"
+ datasets[idx]
+ "_2/"
+ datasets[idx]
+ "_r_"
+ str(ratio)
+ ".csv"
)
num_of_buckets = 100
(
disparity,
disparity_original,
ranking,
theta,
duration,
) = find_fair_ranking(path, n_vectors[idx], columns[idx], sensitive_attrs[idx], num_of_buckets)
# print("Disparity:", disparity)
# print("Original Disparity:", disparity_original)
preprocessing_time.append(duration)
disparities_after.append(disparity)
disparities_before.append(disparity_original)
dataset = read_file(path, columns[idx])
f = polartoscalar([theta], d)
scores = sorted([score(dataset[i], f, d) for i in range(len(dataset))])
query_time = []
for q in queries:
start = timeit.default_timer()
query(q, f, scores, d, num_of_buckets)
stop = timeit.default_timer()
query_time.append(stop - start)
query_times.append(np.mean(query_time))
preprocessing_time_t.append(preprocessing_time)
query_times_t.append(query_times)
disparities_after_t.append(disparities_after)
disparities_before_t.append(disparities_before)
print("Varying number of vectors (prep time):", np.mean(preprocessing_time_t,axis=0))
print("Varying number of vectors (query time):", np.mean(query_times_t,axis=0))
print("Disparity Before:",np.mean(disparities_before_t,axis=0))
print("Disparity After:",np.mean(disparities_after_t,axis=0))
print("Varying minority ratio (prep time):", preprocessing_time)
print("Varying minority ratio (query time):", query_times)
print("Disparity Before:",disparities_before)
print("Disparity After:",disparities_after)
plot_2(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_ratio_unfairness.png",
ratios,
disparities_before,
disparities_after,
ratios,
"Varying ratio (Disparity Before/After)",
"Ratio",
"Disparity Before/After",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_ratio_prep_time.png",
ratios,
preprocessing_time,
ratios,
"Varying ratio (prep time)",
"Ratio",
"Time (sec)",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_ratio_query_time.png",
ratios,
query_times,
ratios,
"Varying ratio (query time)",
"Ratio",
"Time (sec)",
[5e-7, 15e-7],
)
preprocessing_time = []
query_times = []
disparities_after = []
disparities_before = []
for num_of_buckets in num_of_buckets_list:
print(
"=================",
"number of buckets:",
num_of_buckets,
"=================",
)
path = "real_data/" + datasets[idx] + "/" + datasets[idx] + "_r_0.25.csv"
(
disparity,
disparity_original,
ranking,
theta,
duration,
) = find_fair_ranking(path, n_vectors[idx], columns[idx], sensitive_attrs[idx], num_of_buckets)
print("Disparity:", disparity)
print("Original Disparity:", disparity_original)
preprocessing_time.append(duration)
disparities_after.append(disparity)
disparities_before.append(disparity_original)
dataset = read_file(path, columns[idx])
f = polartoscalar([theta], d)
scores = sorted([score(dataset[i], f, d) for i in range(len(dataset))])
query_time = []
for q in queries:
start = timeit.default_timer()
query(q, f, scores, d, num_of_buckets)
stop = timeit.default_timer()
query_time.append(stop - start)
query_times.append(np.mean(query_time))
print("Varying number of buckets (prep time):", preprocessing_time)
print("Varying number of buckets (query time):", query_times)
print("Disparity Before:",disparities_before)
print("Disparity After:",disparities_after)
plot_2(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_num_of_buckets_unfairness.png",
num_of_buckets_list,
disparities_before,
disparities_after,
num_of_buckets_list,
"Varying number of buckets (Disparity Before/After)",
"Number of buckets",
"Disparity Before/After",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_num_of_buckets_prep_time.png",
num_of_buckets_list,
preprocessing_time,
num_of_buckets_list,
"Varying number of buckets (prep time)",
"Number of buckets",
"Time (sec)",
)
plot(
"plots/ranking_sampled_vector/" + datasets[idx] + "/varying_num_of_buckets_query_time.png",
num_of_buckets_list,
query_times,
num_of_buckets_list,
"Varying number of buckets (query time)",
"Number of buckets",
"Time (sec)",
[5e-7, 15e-7],
)
# preprocessing_time_t = []
# query_times_t = []
# disparities_after_t = []
# disparities_before_t = []
# for n_vector in n_vectors:
# preprocessing_time = []
# query_times = []
# disparities_after = []
# disparities_before = []
# for i in range(1):
# print("=================", "n_vector:", n_vector, "=================")
# path = (
# "real_data/"
# + datasets[idx]
# + "/"
# + datasets[idx]
# + "_f_0.2.csv"
# )
# n = pd.read_csv(path).shape[0]
# sex=pd.read_csv(path)[sensitive_attrs[idx]].to_list()
# num_of_buckets = 100
# (
# disparity,
# disparity_original,
# ranking,
# theta,
# duration,
# ) = find_fair_ranking(path, n_vector, columns[idx], sensitive_attrs[idx], num_of_buckets)
# out=[0 if sex[val]=="P1_003N" else 1 for val in ranking]
# with open('ordering_'+datasets[idx]+'.pkl', 'wb') as f:
# pickle.dump(out, f)
# disparities_after.append(disparity)
# disparities_before.append(disparity_original)
# preprocessing_time.append(duration)
# query_time = []
# dataset = read_file(path, columns[idx])
# f = polartoscalar([theta], d)
# scores = sorted([score(dataset[i], f, d) for i in range(len(dataset))])
# for q in queries:
# start = timeit.default_timer()
# query(q, f, scores, d, num_of_buckets)
# stop = timeit.default_timer()
# query_time.append(stop - start)
# query_times.append(np.mean(query_time))
# preprocessing_time_t.append(preprocessing_time)
# query_times_t.append(query_times)
# disparities_after_t.append(disparities_after)
# disparities_before_t.append(disparities_before)
# print("Varying number of vectors (prep time):", np.mean(preprocessing_time_t,axis=1))
# print("Varying number of vectors (query time):", np.mean(query_times_t,axis=1))
# print("Disparity Before:",np.mean(disparities_before_t,axis=1))
# print("Disparity After:",np.mean(disparities_after_t,axis=1))
print(
"###############################################################################################################"
)