-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathTable_2_3_Run.py
178 lines (161 loc) · 8 KB
/
Table_2_3_Run.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
177
178
from sklearn.model_selection import StratifiedKFold
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import accuracy_score
from sklearn.svm import LinearSVC #Linear SVM for feture maps
from sklearn.svm import SVC #Dual SVM for Kernels
from time import time
import numpy as np
import FeatureMaps as maps
import DataReader as DR
#all datasets
datasets=[DR.Splice, DR.Wilt,DR.Guide1, DR.Spambase, DR.Phoneme, DR. Magic, DR.Adult ]
#all dataset names
data_names=['Splice','Wilt', 'Guide 1', 'Spambase', 'Phoneme','Magic','Adult' ]
#set mapping funtions
mapping_functions=[maps.linear, maps.phi_p_1,maps.phi_p_1,maps.phi_p_d,maps.phi_p_d]
#set p values according to the mapping functions in respected order in given Table 1
#p value of the linear kernel is used as a dummy variable and others in respected order in given Table 1
p_vals= [0,1,2,1,2]
#set mapping function an kernel names in respected order in given Table 1
kernel_names=['LIN',r'$\phi_{1,1}$',r'$\phi_{2,1}$',r'$\phi_{1,d}$',r'$\phi_{2,d}$','POL','RBF']
#set of POL kernel parameters
d_params = [2, 3, 4]
#set of RBF kernel parameters
g_params = np.power(10.0, range(-5, 5))
#set of C parameters
c_params = np.power(10.0, range(-5, 5))
#random state for splitting
random_state=42
# a dictionary to store all results
results={'Dataset':[], 'Training Acc.':[], 'Test Acc.':[], 'Training Time':[]}
for i in range(len(datasets)):
results['Dataset'].append(data_names[i])
X_train, X_test, y_train, y_test= datasets[i]()
#set dictionaries to store performance metrics
over_all_performance_metrics= {'Kernel': [], 'Training Acc': [], 'Test Acc': [], 'Training Time': []}
#2-fold cross validation object
inner_cv = StratifiedKFold(n_splits=2, shuffle=True, random_state=random_state)
for m in range(len(mapping_functions)):
over_all_performance_metrics['Kernel'].append(kernel_names[m])
acc_by_param = []
#begin: grid search
for ci in c_params:
all_acc = []
#begin: two-fold
for inner_train_index, inner_test_index in inner_cv.split(X_train, y_train):
#begin: scaling
scaler = StandardScaler()
X_inner_train = scaler.fit_transform(X_train[inner_train_index])
X_inner_test = scaler.transform(X_train[inner_test_index])
#end: scaling
y_inner_train, y_inner_test = y_train[inner_train_index], y_train[inner_test_index]
clf = LinearSVC(C=ci, dual=False).fit(mapping_functions[m](X_inner_train, p=p_vals[m]), y_inner_train)
all_acc.append(accuracy_score(y_inner_test, clf.predict(mapping_functions[m](X_inner_test, p=p_vals[m]))))
#end: two-fold
acc_by_param.append(np.mean(all_acc))
#end: grid search
#get best hyperparameters
best_c = c_params[np.argmax(acc_by_param)]
scaler = StandardScaler()
#being: scaling
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled=scaler.transform(X_test)
#end: scaling
s = time()
XD_train = mapping_functions[m](X_train_scaled,p=p_vals[m])
clf = LinearSVC(C=best_c, dual=False).fit(XD_train, y_train)
over_all_performance_metrics['Training Time'].append( round(time() - s, 4))
over_all_performance_metrics['Training Acc'].append(round(100*accuracy_score(y_train, clf.predict(XD_train)),2))
over_all_performance_metrics['Test Acc'].append(round(100*accuracy_score(y_test, clf.predict(mapping_functions[m](X_test_scaled, p=p_vals[m]))),2))
over_all_performance_metrics['Kernel'].append('POL')
acc_by_param = {'ci': [], 'di': [], 'inner_test_acc': []}
#begin: grid search
for ci in c_params:
for di in d_params:
all_acc = []
#begin: two-fold
for inner_train_index, inner_test_index in inner_cv.split(X_train, y_train):
#begin: scaling
scaler = StandardScaler()
X_inner_train = scaler.fit_transform(X_train[inner_train_index])
X_inner_test = scaler.transform(X_train[inner_test_index])
#end: scaling
y_inner_train, y_inner_test = y_train[inner_train_index], y_train[inner_test_index]
clf = SVC(kernel='poly', C=ci, degree=di).fit(X_inner_train, y_inner_train)
all_acc.append(accuracy_score(y_inner_test, clf.predict(X_inner_test)))
#end: two fold
acc_by_param['ci'].append(ci)
acc_by_param['di'].append(di)
acc_by_param['inner_test_acc'].append(np.mean(all_acc))
#end: grid search
#get best hyperparameters
best_ind = np.argmax(acc_by_param['inner_test_acc'])
best_c, best_d = acc_by_param['ci'][best_ind], acc_by_param['di'][best_ind]
#begin: scaling
scaler = StandardScaler()
X_train_scale = scaler.fit_transform(X_train)
X_test_scale = scaler.transform(X_test)
#end: scaling
s = time()
clf = SVC(kernel='poly', C=best_c, degree=best_d).fit(X_train_scale, y_train)
over_all_performance_metrics['Training Time'].append(round(time() - s, 4))
over_all_performance_metrics['Training Acc'].append(round(100*accuracy_score(y_train, clf.predict(X_train_scale ),2)))
over_all_performance_metrics['Test Acc'].append(round(100*accuracy_score(y_test, clf.predict(X_test_scale),2)))
over_all_performance_metrics['Kernel'].append('RBF')
acc_by_param = {'ci': [], 'gi': [], 'inner_test_acc': []}
#begin: grid search
for ci in c_params:
for gi in g_params:
all_acc = []
#begin: two-fold
for inner_train_index, inner_test_index in inner_cv.split(X_train, y_train):
#begin: scaling
scaler = StandardScaler()
X_inner_train = scaler.fit_transform(X_train[inner_train_index])
X_inner_test = scaler.transform(X_train[inner_test_index])
#end: scaling
y_inner_train, y_inner_test = y_train[inner_train_index], y_train[inner_test_index]
clf = SVC(kernel='rbf', C=ci, gamma=gi).fit(X_inner_train, y_inner_train)
all_acc.append(accuracy_score(y_inner_test, clf.predict(X_inner_test)))
#end: two-fold
acc_by_param['ci'].append(ci)
acc_by_param['gi'].append(gi)
acc_by_param['inner_test_acc'].append(np.mean(all_acc))
#end: grid search
#get best hyperparameters
best_ind = np.argmax(acc_by_param['inner_test_acc'])
best_c, best_g = acc_by_param['ci'][best_ind], acc_by_param['gi'][best_ind]
#begin: scaling
scaler = StandardScaler()
X_train_scale = scaler.fit_transform(X_train)
X_test_scale = scaler.transform(X_test)
#end: scaling
s = time()
clf = SVC(kernel='rbf', C=best_c, gamma=best_g).fit(X_train_scale, y_train)
over_all_performance_metrics['Training Time'].append(round(time() - s, 4))
over_all_performance_metrics['Training Acc'].append(round(100 * accuracy_score(y_train, clf.predict(X_train_scale), 2)))
over_all_performance_metrics['Test Acc'].append(round(100 * accuracy_score(y_test, clf.predict(X_test_scale), 2)))
results['Training Time'].append(over_all_performance_metrics['Training Time'])
results['Training Acc.'].append(over_all_performance_metrics['Training Acc'])
results['Test Acc.'].append(over_all_performance_metrics['Test Acc'])
print('***** Table 2: Test Accuracies')
n_datasets=len(datasets)
n_kenels=len(kernel_names)
for d in range(n_datasets):
print()
print('***'+ results['Dataset'][d])
temp=''
for k in range(n_kenels):
temp+=kernel_names[k] +'\t'+ str(results['Test Acc.'][d][k])+ '\t'+ '\t'
print(temp)
print()
print('***** Table 3: Training Times in Seconds')
n_datasets=len(datasets)
n_kenels=len(kernel_names)
for d in range(n_datasets):
print()
print('***'+ results['Dataset'][d])
temp=''
for k in range(n_kenels):
temp+=kernel_names[k] +'\t'+ str(results['Training Time'][d][k])+ '\t'+ '\t'
print(temp)