-
Notifications
You must be signed in to change notification settings - Fork 1
/
model.py
167 lines (140 loc) · 6.58 KB
/
model.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
import numpy as np
import random
import os
import torch
import torch.nn as nn
from torch import Tensor
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader, SubsetRandomSampler, ConcatDataset
import torch.nn.functional as F
from torch.cuda.amp.grad_scaler import GradScaler
from torch.cuda.amp import autocast
from torchsummary import summary
from sklearn.metrics import r2_score
from ray import tune
import json
import itertools
from itertools import groupby
import gzip
from io import BytesIO
from time import time
import matplotlib.pyplot as plt
import pyBigWig
from scipy.sparse import csc_matrix
import math
class upd_GELU(nn.Module):
def __init__(self):
super(upd_GELU, self).__init__()
self.constant_param = nn.Parameter(torch.Tensor([1.702]))
self.sig = nn.Sigmoid()
def forward(self, input: Tensor) -> Tensor:
outval = torch.mul(self.sig(torch.mul(self.constant_param, input)), input)
return outval
def ones_(tensor: Tensor) -> Tensor:
return torch.ones_like(tensor)
def zeros_(tensor: Tensor) -> Tensor:
return torch.zeros_like(tensor)
class BasenjiModel(nn.Module):
def __init__(self, num_targets, n_channel=4, max_len=128,
conv1kc=64, conv1ks=15, conv1st=1, conv1pd=7, pool1ks=8, pool1st=1 , pdrop1=0.4, #conv_block_1 parameters
conv2kc=64, conv2ks=5, conv2st=1, conv2pd=3, pool2ks=4 , pool2st=1, pdrop2=0.4, #conv_block_2 parameters
conv3kc=round(64*1.125), conv3ks=5, conv3st=1, conv3pd=3, pool3ks=4 , pool3st=1, pdrop3=0.4, #conv_block_2 parameters
convdc = 6, convdkc=32 , convdks=3, debug=False):
super(BasenjiModel, self).__init__()
self.convdc = convdc
self.debug = debug
self.num_targets = num_targets
## CNN + dilated CNN
self.conv_block_1 = nn.Sequential(
upd_GELU(),
nn.Conv1d(n_channel, conv1kc, kernel_size=conv1ks, stride=conv1st, padding=conv1pd, bias=False),
nn.BatchNorm1d(conv1kc, momentum=0.9, affine=True),
nn.MaxPool1d(kernel_size=pool1ks),
nn.Dropout(p=0.2))
self.conv_block_2 = nn.Sequential(
upd_GELU(),
nn.Conv1d(conv1kc, conv2kc, kernel_size=conv2ks, stride=conv2st, padding=conv2pd, bias=False),
nn.BatchNorm1d(conv2kc, momentum=0.9, affine=True),
nn.MaxPool1d(kernel_size=pool2ks),
nn.Dropout(p=0.2))
self.conv_block_3 = nn.Sequential(
upd_GELU(),
nn.Conv1d(conv2kc, round(conv2kc*1.125), kernel_size=conv3ks, stride=conv3st, padding=conv3pd, bias=False),
nn.BatchNorm1d(conv3kc, momentum=0.9, affine=True),
nn.MaxPool1d(kernel_size=pool3ks),
nn.Dropout(p=0.2))
self.dilations = nn.ModuleList()
for i in range(convdc):
padding = 2**(i)
self.dilations.append(nn.Sequential(
upd_GELU(),
nn.Conv1d(conv3kc, 32, kernel_size=3, padding=padding, dilation=2**i, bias=False),
nn.BatchNorm1d(32, momentum=0.9, affine=True),
upd_GELU(),
nn.Conv1d(32, 72, kernel_size=1, padding=0, bias=False),
nn.BatchNorm1d(72, momentum=0.9, affine=True),
nn.Dropout(p=0.25)))
self.conv_block_4 = nn.Sequential(
upd_GELU(),
nn.Conv1d(72, 64, kernel_size=1, stride=1, padding=0, bias=False),
nn.BatchNorm1d(64, momentum=0.9, affine=True),
nn.Dropout(p=0.1))
self.conv_block_5 = nn.Sequential(
upd_GELU(),
nn.Linear(64, self.num_targets, bias=True),
nn.Softplus(beta=1, threshold=1000))
self.conv_block_1[1].weight.data = self.truncated_normal(self.conv_block_1[1].weight, 0.0, np.sqrt(2/60)) #4
self.conv_block_2[1].weight.data = self.truncated_normal(self.conv_block_2[1].weight, 0.0, np.sqrt(2/322)) # conv1kc
self.conv_block_3[1].weight.data = self.truncated_normal(self.conv_block_3[1].weight, 0.0, np.sqrt(2/322)) # conv1kc
self.conv_block_4[1].weight.data = self.truncated_normal(self.conv_block_4[1].weight, 0.0, np.sqrt(2/72)) # 72
self.conv_block_5[1].weight.data = self.truncated_normal(self.conv_block_5[1].weight, 0.0, np.sqrt(2/64)) # 64
self.conv_block_1[2].weight.data = ones_(self.conv_block_1[2].weight)
self.conv_block_2[2].weight.data = ones_(self.conv_block_2[2].weight)
self.conv_block_3[2].weight.data = ones_(self.conv_block_3[2].weight)
self.conv_block_4[2].weight.data = ones_(self.conv_block_4[2].weight)
for i in range(convdc):
self.dilations[i][1].weight.data = self.truncated_normal(self.dilations[i][1].weight, 0.0, np.sqrt(2/218)) # 72
self.dilations[i][-2].weight.data = self.truncated_normal(self.dilations[i][-2].weight, 0.0, np.sqrt(2/32)) # 32
self.dilations[i][2].weight.data = zeros_(self.dilations[i][2].weight)
self.dilations[i][-2].weight.data = ones_(self.dilations[i][-2].weight)
def truncated_normal(self, t, mean, std):
torch.nn.init.normal_(t, mean, std)
while True:
cond = torch.logical_or(t < (mean - 2.28*std), t > (mean + 2.28*std))
if not torch.sum(cond):
break
t = torch.where(cond, torch.nn.init.normal_(torch.ones(t.shape), mean=mean, std=std), t)
return t
def forward(self, seq):
if self.debug:
print (seq.shape)
seq = self.conv_block_1(seq)
if self.debug:
print ('conv1', seq.shape)
seq = self.conv_block_2(seq)
if self.debug:
print ('conv2', seq.shape)
seq = self.conv_block_3(seq)
if self.debug:
print ('conv3', seq.shape)
for i in range(self.convdc):
if i == 0:
y = self.dilations[i](seq)
if i >= 1:
y = y.add(self.dilations[i](seq))
if self.debug:
print ('dil', i, self.dilations[i](seq).shape)
if self.debug:
print ('y', y.shape)
res = self.conv_block_4(y)
if self.debug:
print ('res', res.shape)
res_lin = res.transpose(1, 2)
if self.debug:
print ('res_lin', res_lin.shape)
res = self.conv_block_5(res_lin)
if self.debug:
print ('res', res.shape)
return res
def compile(self, device='cpu'):
self.to(device)