ml-bcell-epitope/script5.py at master · johncava/ml-bcell-epitope · 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
from __future__ import division
from utility import *
import torch
import random
from torch.autograd import Variable

test_data = load_test('SEQ194.txt')
#print test[0]

protvec = initialize()

# Positive and Negative Datasets
pos_data = []
neg_data = []

# Read in the Positive Dataset
with open('pos.data') as f:
    for line in f:
        line = line.rstrip()
        p = embedding(protvec,line)
        pos_data.append([p,1.0])

# Read in the Negative Dataset
with open('neg.data') as f:
    for line in f:
        n = []
        line = line.rstrip()
        n = embedding(protvec, line)
        neg_data.append([n,0.0])

# Data preparation
data = pos_data + neg_data
data = np.array(data)
np.random.shuffle(data)

data = data.tolist()
x,y = data[0][0], data[0][1]
#print x
print y
#input = Variable(torch.from_numpy(np.array(x))).view(1,3,100).double()
#print input

# 1D convolution
class Discrim(torch.nn.Module):
    def __init__(self):
        super(Discrim, self).__init__()
        self.c1 = torch.nn.Conv1d(3,20,3).double()
        self.relu = torch.nn.LeakyReLU(0.1)
        self.drop = torch.nn.Dropout()
        self.p1 = torch.nn.MaxPool1d(2)
        self.c2 = torch.nn.Conv1d(20,1,2).double()
        #torch.nn.LeakyReLU(0.1)
        #torch.nn.Dropout()
        self.p2 = torch.nn.MaxPool1d(2)
        self.linear = torch.nn.Linear(24,12).double()
        self.linear2 = torch.nn.Linear(12,12).double()
        self.linear3 = torch.nn.Linear(12,1).double()
	#self.tanh = torch.nn.Tanh()
	self.sigmoid = torch.nn.Sigmoid()

    def forward(self, input):
        x = self.c1(input)
        x = self.relu(x)
        #x = self.drop(x)
        x = self.p1(x)
        x = self.c2(x)
        x = self.relu(x)
        #x = self.drop(x)
        x = self.p2(x)
        x = self.linear(x.view(1,24))
        #x = self.drop(x)
        x = self.linear2(x)
	x = self.linear3(x)
        return self.sigmoid(x)

loss_fn = torch.nn.MSELoss(size_average=True)

learning_rate = 1e-4

model = Discrim()

#print len(data) --> 633

optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)

#print data[0]

for epoch in xrange(5):
	for index in xrange(633):
		#print index
		train = data[index]
		x , y = train[0] , train[1]
		#print x,y
		#break
		x = Variable(torch.from_numpy(np.array(x)), requires_grad = False).view(1,3,100).double()

		inpt_train_y = torch.from_numpy(np.array([[y]]))
		inpt_train_y = inpt_train_y.double()
		inpt_train_y = Variable(inpt_train_y, requires_grad=False)
		#print inpt_train_y
		y_pred = model(x)
		loss = loss_fn(y_pred, inpt_train_y)
		if index%100 == 0:
			print loss
		optimizer.zero_grad()
		loss.backward()
		optimizer.step()

# Testing

window = 20
for test in test_data:
	sequence, label = test[1], test[2]
	# Create window lists
	prediction = [[]] * len(sequence)
	indices = range(len(sequence))
	indices_list = []
	for i in xrange(len(sequence) - window):
		a = indices[i:i+window]
		indices_list.append(a)
		# Predict
		embed = embedding(protvec, sequence[i:i+window])
		#print embed
		x = Variable(torch.from_numpy(np.array(embed)), requires_grad = False).view(1,3,100).double()
		pred = model(x)
		#print pred[0][0].data.numpy()
		for index in a:
			prediction[index].append(pred[0][0].data.numpy().tolist())
		break
	for index, p in enumerate(prediction):
		prediction[index] = sum(p) / float(len(p))
	for i in prediction:
		if i > 0.5:
			print 'hi'
	break