COS424/feature_selection.py at master · binarybin/COS424 · 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
"""
Usage: python feature_selection.py auto <C> <kernel=linear, rbf, poly, sigmoid> <gamma>
"""


import pandas as pd
import numpy as np
from random import sample
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer
from sklearn.naive_bayes import MultinomialNB, GaussianNB, BernoulliNB
from sklearn.svm import SVC, LinearSVC
from sklearn.metrics import classification_report, f1_score, accuracy_score, confusion_matrix, roc_curve, auc
from sklearn.pipeline import Pipeline
from sklearn.grid_search import GridSearchCV
from sklearn.cross_validation import StratifiedKFold, cross_val_score, train_test_split
from sklearn.tree import DecisionTreeClassifier
from sklearn.learning_curve import learning_curve
import cPickle as pickle
from sys import argv

C = int(argv[2])
kernel = argv[3]
gamma = float(argv[4])


bid=pd.read_csv('bids.csv')
train=pd.read_csv('train.csv')
print "Data loaded"


def extract_feature(df):
    gb=df.groupby('bidder_id')
    #number of countries
    country_count=gb['country'].apply(pd.value_counts).count(level=0)
    #of unique device
    device_count=gb['device'].apply(pd.value_counts).count(level=0)
    # of unique ip
    ip_count=gb['ip'].apply(pd.value_counts).count(level=0)
    # of unique auction
    auction_count=gb['auction'].apply(pd.value_counts).count(level=0)
    # of unique url
    url_count=gb['url'].apply(pd.value_counts).count(level=0)
    # of transitions
    time_count=gb['time'].apply(pd.value_counts).count(level=0)
    #time interval **** zero value represents that only one transition has happened ****
    grouped=[gb.get_group(x) for x in gb.groups]
    rows_list=[]
    for i in range(0,len(grouped)):
        dict1={}
        dict1.update({'bidder_id':grouped[i]['bidder_id'].iloc[0],'bidderID':grouped[i]['bidder_id'].iloc[0],'time interval':grouped[i]['time'].iloc[len(grouped[i].index)-1]-grouped[i]['time'].iloc[0]})
        rows_list.append(dict1)
    time = pd.DataFrame(rows_list)
    time_int=time.set_index('bidder_id')
    #average transition time: time interval/transitions
    pieces = [country_count,device_count,ip_count,auction_count,url_count,time_count]
    concatenated = pd.concat(pieces,axis=1,keys=['country', 'device','ip','auction','url','transitions'])
    concatenated = pd.concat([concatenated,time_int],axis=1)
    return concatenated

try:
    with open("features", 'rb') as fp:
        a = pickle.load(fp)
    print "Feature loaded!"

except:
    print "Feature loading failed, re-extracting!"
    a=extract_feature(bid)
    with open("features", 'wb') as fp:
        pickle.dump(a, fp)
    print "Feature extracted!"


train_size = 1200
rindex =  np.array(sample(xrange(len(train)), train_size))
real_train = train.ix[rindex]
real_test = train[~train.isin(real_train).all(1)]
trainable = a[a.bidderID.isin(real_train.bidder_id)]
testable = a[a.bidderID.isin(real_test.bidder_id)]

train_df = pd.concat([real_train.set_index('bidder_id'), trainable], axis=1)
train_df = train_df[np.isfinite(train_df['ip'])]
train_df = train_df.reset_index().drop("index", 1).drop("payment_account", 1).drop("address", 1).drop("bidderID",1)

test_df = pd.concat([real_test.set_index('bidder_id'), testable], axis=1)
test_df = test_df[np.isfinite(test_df['ip'])]
test_df = test_df.reset_index().drop("index", 1).drop("payment_account", 1).drop("address", 1).drop("bidderID",1)


train_x, train_y = train_df.values[:,1:], train_df.values[:,0].astype(int)
test_x, test_y = test_df.values[:,1:], test_df.values[:,0].astype(int)

print "Train data selected"

def Predict(model, train_x, train_y, test_x, test_y):
    detector = model.fit(train_x, train_y)
    predictions = detector.predict(test_x)
    print 'accuracy', accuracy_score(predictions, test_y)
    print 'confusion matrix\n', confusion_matrix(test_y, predictions)
    print '(row=expected, col=predicted)'
    return accuracy_score, confusion_matrix

def ROC(model, train_x, train_y, test_x, test_y):
    detector = model.fit(train_x, train_y)
    y_score = detector.predict_proba(test_x)
    fpr, tpr, _ = roc_curve(test_y, y_score[:,1])
    roc_auc = auc(fpr, tpr)
    return fpr, tpr, roc_auc

if kernel == "linear":
    model = SVC(C = C,  kernel = kernel, probability = True)
else:
    model = SVC(C = C,  kernel = kernel, gamma = gamma, probability = True)

if argv[1] == "auto":
    filename = "SVM_"+kernel+"_C_"+str(C)+"_gamma_"+str(gamma)
else:
    filename = argv[1]

Prediction_result = Predict(model, train_x, train_y, test_x, test_y)

with open(filename+"PRED", 'wb') as fp:
    pickle.dump(Prediction_result, fp)
print "prediction finished"


ROC_result = ROC(model, train_x, train_y, test_x, test_y)

with open(filename+"ROC", 'wb') as fp:
    pickle.dump(ROC_result, fp)
print "ROC finished"

"""
# In[ ]:

ROC(SVC(C = 1,  kernel = "linear", probability = True), train_x, train_y, test_x, test_y)


# In[ ]:

ROC(SVC(C = 1000,  kernel = "rbf", gamma = 0.001, probability = True), train_x, train_y, test_x, test_y)


# In[ ]:

"""