/
predict.py
176 lines (138 loc) · 5.54 KB
/
predict.py
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
"""
analysis.py
Facebook Recruiting IV: Human or Robot?
author: Yusuke Sakamoto
"""
import numpy as np
import pandas as pd
import time
from sklearn.metrics import roc_curve, auc
import matplotlib.pyplot as plt
from etc import predict_usample
start_time = time.time()
print "Loading postprocessed data files..."
humanfile = 'data/human_info.csv'
botsfile = 'data/bots_info.csv'
testfile = 'data/test_info.csv'
human_info = pd.read_csv(humanfile, index_col=0)
bots_info = pd.read_csv(botsfile, index_col=0)
test_info = pd.read_csv(testfile, index_col=0)
num_human = human_info.shape[0]
num_bots = bots_info.shape[0]
num_test = test_info.shape[0]
# number of bids per auction data (descending)
bbba = pd.read_csv('data/bots_bids_by_aucs.csv', index_col=0)
hbba = pd.read_csv('data/human_bids_by_aucs.csv', index_col=0)
tbba = pd.read_csv('data/test_bids_by_aucs.csv', index_col=0)
# take only some data from auc_count
max_auc_count = 1000
max_auc_count = min([bbba.shape[1], hbba.shape[1], tbba.shape[1],
max_auc_count])
####
# train_ids = train_info['bidder_id']
test_ids = test_info.index
############################################################################
# Data dropping/appending
############################################################################
# append num_bids_by_auc_ columns
if max_auc_count > 0:
# max_auc_count = 100
hbba.fillna(0)
bbba.fillna(0)
tbba.fillna(0)
human_info = pd.concat([human_info, hbba.iloc[:, :max_auc_count]], axis=1)
bots_info = pd.concat([bots_info, bbba.iloc[:, :max_auc_count]], axis=1)
test_info = pd.concat([test_info, tbba.iloc[:, :max_auc_count]], axis=1)
# drop num item labels
# columns_dropped = [u'num_merchandise', u'num_devices', u'num_countries', u'num_ips',
# u'num_urls']
columns_dropped = [u'num_merchandise']
human_info.drop(columns_dropped, axis=1, inplace=True)
bots_info.drop(columns_dropped, axis=1, inplace=True)
test_info.drop(columns_dropped, axis=1, inplace=True)
# drop merchandise dummy variables
for key in human_info.keys():
if 'merchandise' in key:
human_info.drop([key], axis=1, inplace=True)
bots_info.drop([key], axis=1, inplace=True)
test_info.drop([key], axis=1, inplace=True)
human_info.sort(axis=1)
bots_info.sort(axis=1)
test_info.sort(axis=1)
############################################################################
# Predicting: Random Forest with bagging
############################################################################
# bagging with bootstrap
y_valids = []
roc_auc_mean = []
roc_auc_std = []
bots_rate_mean = []
bots_rate_std = []
specificity_mean = []
specificity_std = []
score_mean = []
score_std = []
holdout = 0.2
# ks = range(1,18,4)
ks = range(20, 21)
for k in ks:
num_sim = 5
y_probas = []
roc_aucs = []
tprs = []
scovs = []
for i in range(num_sim):
np.random.seed(int(time.time() * 1000 % 4294967295))
y_proba, y_pred, train_proba, train_pred, roc_auc, tpr, scov\
= predict_usample(num_human, num_bots, human_info,
bots_info, test_info, holdout=holdout,
multiplicity=k, plot_roc=True)
y_probas.append(y_proba[:, 1]) # gather the bot probabilities
roc_aucs.append(roc_auc)
tprs.append(tpr)
scovs.append(scov)
tprs = np.array(tprs)
specificity_mean.append(tprs.mean())
specificity_std.append(tprs.std())
scovs = np.array(scovs)
score_mean.append(scovs.mean())
score_std.append(scovs.std())
# postprocessing
y_probas = np.array(y_probas)
y_proba_ave = y_probas.T.mean(axis=1)
bots_rates = np.sum(y_probas > 0.5, axis=1) / \
np.array(map(len, y_probas), dtype=float)
bots_rate_mean.append(bots_rates.mean())
bots_rate_std.append(bots_rates.std())
roc_aucs = np.array(roc_aucs)
roc_auc_mean.append(roc_aucs.mean())
roc_auc_std.append(roc_aucs.std())
# print "k: ", k
# print "bots proba for test set: ", brs.mean()
np.set_printoptions(suppress=True, precision=3)
print "CV result:"
cv_result = np.round(np.array([ks[0], roc_auc_mean[0], roc_auc_std[0],
bots_rate_mean[0], bots_rate_std[0],
specificity_mean[0],
specificity_std[0], score_mean[0],
score_std[0]]), 3)
cv_scores = ['multiplicity', 'roc_auc: mean', 'roc_auc: std',
'bots_rate: mean', 'bote_rate: std', 'specificity: mean',
'specificity: std', 'accuracy: mean', 'accuracy: std']
print pd.DataFrame(cv_result, index=cv_scores)
# 70 bidders in test.csv do not have any data in bids.csv. Thus they
# are not included in analysis/prediction, but they need to be
# appended in the submission. The prediction of these bidders do not matter.
test_ids_all = pd.read_csv('data/test.csv')['bidder_id']
test_ids_append = list(
set(test_ids_all.values).difference(set(test_ids.values)))
submission_append = pd.DataFrame(np.zeros(len(test_ids_append)),
index=test_ids_append, columns=['prediction'])
# Make as submission file!
submission = pd.DataFrame(y_proba_ave, index=test_ids, columns=['prediction'])
submission = pd.concat([submission, submission_append], axis=0)
submission.to_csv('data/submission.csv', index_label='bidder_id')
print "bots proba for train set:", num_bots / float(num_human + num_bots)
print "bots proba for test set: ", sum(y_proba_ave > 0.5) / float(len(y_proba_ave))
end_time = time.time()
print "Time elapsed: %.2f" %(end_time-start_time)