Kernel-Methods-For-Machine-Learning-MVA-Challenge/start.py at master · PirashanthR/Kernel-Methods-For-Machine-Learning-MVA-Challenge · 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
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
194
195
196
197
'''
Start -- Kernel Methods for Machine Learning 2017-2018 -- RATNAMOGAN Pirashanth -- SAYEM Othmane
Required file that computes our final submission
'''

from Kernel.SpectrumKernel import AllPossibleCombinationlist
from Kernel.SpectrumKernel import CreateHistogramMismatchSeq,compute_idf
from Classifiers.KernelLogisticRegression import KernelLogisticRegression
from Kernel.SpectrumKernel import histogram_kernel
from Tools.Utils import X_train_matrix_0,X_train_matrix_1,X_train_matrix_2,\
X_train_0,X_train_1,X_train_2,Y_train_0,Y_train_1,Y_train_2,X_test_0\
,X_test_1,X_test_2,X_test_matrix_0,X_test_matrix_1,X_test_matrix_2
from Classifiers.KernelSVM import SVMC
import numpy as np
from Tools.Utils import accuracy_score
from Tools.Utils import train_test_split

nngram = 7 #param
list_all_combin_DNA = AllPossibleCombinationlist(['A','C','G','T'],nngram)
X_train_histo_0 = np.empty([len(X_train_0),len(list_all_combin_DNA)])
X_test_histo_0 = np.empty([len(X_test_0),len(list_all_combin_DNA)])
X_train_histo_1 = np.empty([len(X_train_1),len(list_all_combin_DNA)])
X_test_histo_1 = np.empty([len(X_test_1),len(list_all_combin_DNA)])
X_train_histo_2 = np.empty([len(X_train_2),len(list_all_combin_DNA)])
X_test_histo_2 = np.empty([len(X_test_2),len(list_all_combin_DNA)])

#################Read Data#######################

for i in range(len(X_train_0)):
    X_train_histo_0[i,:] = CreateHistogramMismatchSeq(X_train_0[i],list_all_combin_DNA,nngram)

for j in range(len(X_test_0)):
    X_test_histo_0[j,:] = CreateHistogramMismatchSeq(X_test_0[j],list_all_combin_DNA,nngram)

for i in range(len(X_train_1)):
    X_train_histo_1[i,:] = CreateHistogramMismatchSeq(X_train_1[i],list_all_combin_DNA,nngram)

for j in range(len(X_test_1)):
    X_test_histo_1[j,:] = CreateHistogramMismatchSeq(X_test_1[j],list_all_combin_DNA,nngram)


for i in range(len(X_train_2)):
    X_train_histo_2[i,:] = CreateHistogramMismatchSeq(X_train_2[i],list_all_combin_DNA,nngram)

for j in range(len(X_test_2)):
    X_test_histo_2[j,:] = CreateHistogramMismatchSeq(X_test_2[j],list_all_combin_DNA,nngram)


X_train_split_0 = X_train_histo_0
y_train_split_0 = Y_train_0
y_train_split_0[y_train_split_0==0]=-1

X_train_split_1 = X_train_histo_1
y_train_split_1 = Y_train_1
y_train_split_1[y_train_split_1==0]=-1

X_train_split_2 = X_train_histo_2
y_train_split_2 = Y_train_2
y_train_split_2[y_train_split_2==0]=-1

#################Compute the train gram matrices#######################
gram_train_multi_proc = histogram_kernel(X_train_split_0,n_proc=1)
gram_train_multi_proc_1 = histogram_kernel(X_train_split_1,n_proc=1)
gram_train_multi_proc_2 = histogram_kernel(X_train_split_2,n_proc=1)

#################Compute test gram matrices#######################
gram_test_final_0 =  histogram_kernel(X_train_split_0,X_test_histo_0,n_proc=1)
gram_test_final_1 =  histogram_kernel(X_train_split_1,X_test_histo_1,n_proc=1)
gram_test_final_2 =  histogram_kernel(X_train_split_2,X_test_histo_2,n_proc=1)


c = 0.5
sv = 1e-4
lambda_log_reg = 1
tolerance = 0.001
list_of_prediction_test_0= []

for i in range(11):
    if (i==0)or(i==11):
        list_train = list(range(2000))
        list_val = []
    else:
        test_size = 1/10
        list_train,list_val = train_test_split(list(range(2000)),test_size=test_size)
    gram_train_multi_proc_cur = (gram_train_multi_proc[list_train,:][:,list_train])
    gram_val_multi_proc = (gram_train_multi_proc[list_train,:][:,list_val])
    y_train_split_0_cur = y_train_split_0[list_train]
    y_val_split_0 = y_train_split_0[list_val]

    if i<100:
        print('opt c sv',c,sv)
        svm_test = SVMC(c= c,min_sv = sv)
        svm_test.fit(gram_train_multi_proc_cur,y_train_split_0_cur)
        y_train_pred = svm_test.predict_class(gram_val_multi_proc).reshape(-1)
        print('0 Precision SVM=',accuracy_score(y_val_split_0.reshape(-1),y_train_pred))
        y_test_pred_0 = svm_test.predict_class(gram_test_final_0[list_train,:])
        y_test_pred_0[y_test_pred_0==-1]=0

    else:
        log_reg = KernelLogisticRegression()
        log_reg.fit(gram_train_multi_proc_cur,y_train_split_0_cur,lambda_regularisation=lambda_log_reg,tolerance=tolerance)
        y_train_pred = log_reg.predict_class(gram_val_multi_proc).reshape(-1)
        print('0 Precision Log Reg=',accuracy_score(y_val_split_0.reshape(-1),y_train_pred))
        y_test_pred_0 = log_reg.predict_class(gram_test_final_0[list_train,:])
        y_test_pred_0[y_test_pred_0==-1]=0
    list_of_prediction_test_0.append(y_test_pred_0)


y_pred_0 = np.array(np.array(list_of_prediction_test_0).mean(axis=0).reshape((-1,))>0.5,dtype=int) #average output of LGBM and XGB

c=0.5
list_of_prediction_test_1= []

for i in range(11):
    if (i==0)or(i==11):
        list_train = list(range(2000))
        list_val = []
    else:
        test_size = 1/10
        list_train,list_val = train_test_split(list(range(2000)),test_size=test_size)

    y_train_split_1_cur = y_train_split_1[list_train]
    y_val_split_1 = y_train_split_1[list_val]

    gram_train_multi_proc_1_cur = (gram_train_multi_proc_1[list_train,:][:,list_train])
    gram_val_multi_proc1 = (gram_train_multi_proc_1[list_train,:][:,list_val])

    if i<100:
        print('opt c sv',c,sv)
        svm_test_1 = SVMC(c= c,min_sv = sv)
        svm_test_1.fit(gram_train_multi_proc_1_cur,y_train_split_1_cur)

        y_train_pred = svm_test_1.predict_class(gram_val_multi_proc1).reshape(-1)
        print('1 Precision =',accuracy_score(y_val_split_1.reshape(-1),y_train_pred))
        y_test_pred_1 = svm_test_1.predict_class(gram_test_final_1[list_train,:])
        y_test_pred_1[y_test_pred_1==-1]=0
    else:
        log_reg_1 = KernelLogisticRegression()
        log_reg_1.fit(gram_train_multi_proc_1_cur,y_train_split_1_cur,lambda_regularisation=lambda_log_reg,tolerance=tolerance)
        y_train_pred = log_reg_1.predict_class(gram_val_multi_proc1).reshape(-1)
        print('1 Precision Log Reg=',accuracy_score(y_val_split_1.reshape(-1),y_train_pred))
        y_test_pred_1 = log_reg_1.predict_class(gram_test_final_1[list_train,:])
        y_test_pred_1[y_test_pred_1==-1]=0
    list_of_prediction_test_1.append(y_test_pred_1)


y_pred_1 = np.array(np.array(list_of_prediction_test_1).mean(axis=0).reshape((-1,))>0.5,dtype=int) #average output of LGBM and XGB


c=0.5
sv =1e-4
list_of_prediction_test_2 = []

for i in range(11):
    if (i==0)or(i==11):
        list_train = list(range(2000))
        list_val = []
    else:
        test_size = 1/10
        list_train,list_val = train_test_split(list(range(2000)),test_size=test_size)

    y_train_split_2_cur = y_train_split_2[list_train]
    y_val_split_2 = y_train_split_2[list_val]


    gram_train_multi_proc_2_cur = (gram_train_multi_proc_2[list_train,:][:,list_train])
    gram_val_multi_proc2 = (gram_train_multi_proc_2[list_train,:][:,list_val])

    if i<100:
        print('opt c sv',c,sv)
        svm_test_2 = SVMC(c=c,min_sv = sv)
        svm_test_2.fit(gram_train_multi_proc_2_cur,y_train_split_2_cur)

        y_train_pred = svm_test_2.predict_class(gram_val_multi_proc2).reshape(-1)
        print('2 Precision =',accuracy_score(y_val_split_2.reshape(-1),y_train_pred))
        y_test_pred_2 = svm_test_2.predict_class(gram_test_final_2[list_train,:])
        y_test_pred_2[y_test_pred_2==-1]=0

    else:
        log_reg_2 = KernelLogisticRegression()
        log_reg_2.fit(gram_train_multi_proc_2_cur,y_train_split_2_cur,lambda_regularisation=lambda_log_reg,tolerance=tolerance)
        y_train_pred = log_reg_2.predict_class(gram_val_multi_proc2).reshape(-1)
        print('2 Precision Log Reg=',accuracy_score(y_val_split_2.reshape(-1),y_train_pred))
        y_test_pred_2 = log_reg_2.predict_class(gram_test_final_2[list_train,:])
        y_test_pred_2[y_test_pred_2==-1]=0
    list_of_prediction_test_2.append(y_test_pred_2)

y_pred_2 = np.array(np.array(list_of_prediction_test_2).mean(axis=0).reshape((-1,))>0.5,dtype=int) #average output of LGBM and XGB

y_pred = list(y_pred_0)+list(y_pred_1)+list(y_pred_2)

with open("Yte.csv", 'w') as f:
    f.write('Id,Bound\n')
    for i in range(len(y_pred)):
        f.write(str(i)+','+str(y_pred[i])+'\n')