-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbert_ensemble_train.py
188 lines (157 loc) · 6.41 KB
/
bert_ensemble_train.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
179
180
181
182
183
184
185
186
187
188
from model.bert import BERT
import bert_pytorch
import torch
import torch.nn as nn
from dataset.dataset import BERTDataset
from dataset.vocab import WordVocab
from torch.utils.data import DataLoader
import csv
import random
from torch.utils.data import TensorDataset, DataLoader
import numpy as np
from ENSEMBLE_BERT import DeepSEA_BERT
import os
import argparse
base_dict = {'A':0, 'G':1, 'C':2, 'T':3}
def seq2idxseq(seq, base_dict):
idx_list = [np.zeros(len(seq)) for i in range(len(base_dict))]
for i in range(len(seq)):
idx_list[base_dict[seq[i]]][i] = 1
return idx_list
def tensor_generate(data_set):
x = torch.FloatTensor(np.array([seq2idxseq(x[0], base_dict) for x in data_set])).cuda()
y = torch.FloatTensor(np.array([x[1] for x in data_set])).cuda()
return x,y
def seq_split(seq):
head = list(seq)
return ' '.join(head)
def sentence_processing(sentence1):
out1 = [vocab.sos_index] + vocab.to_seq(sentence1) + [vocab.eos_index]
seg_labels = [1 for x in out1]
return (out1, seg_labels)
def data_generate(data_set):
seq_list= []
label_list = []
target_list = []
for row in data_set:
curr_seq = seq_split(row[0])
bert_seq,bert_label = sentence_processing(curr_seq)
seq_list.append(bert_seq)
label_list.append(bert_label)
target_list.append(row[1])
return torch.LongTensor(seq_list).cuda(),torch.LongTensor(label_list).cuda(),torch.FloatTensor(target_list).cuda()
def training_loop(model, loss, optimizer, epochs,batch_size, saved_path):
train_loader = DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True)
for j in range(epochs):
loss_sum = 0
for i,data in enumerate(train_loader):
if i % 2000 == 0:
print(i)
model.train()
bert_seq,bert_label,sea, target = data
if target.size(0) <= 1:
continue
model.zero_grad()
output = model(bert_seq,bert_label, sea)
lossy = loss(output, target)
lossy.backward()
optimizer.step()
loss_sum = loss_sum + lossy.detach().cpu().numpy()
torch.save(model, os.path.join(saved_path, 'bert.ensemble.ep'+ str(j+1)))
print( "Epochs %i; Loss %f, Validation Accuracy %f" %(j+1, loss_sum, test_model(valid_dataset, model, batch_size, th=0.5)))
def test_model(xy_set, model, batch_size, th=0.5):
model.eval()
data_loader = DataLoader(dataset=xy_set,batch_size=batch_size,shuffle=True)
hit = 0
total = 0
for i,data in enumerate(data_loader):
bert_seq,bert_label,sea,target = data
out = model(bert_seq,bert_label, sea)
out = out.detach().cpu().numpy()
label = target.cpu().numpy()
for j in range(len(label)):
curr = 0
if out[j] > th:
curr = 1
if label[j] == curr:
hit = hit + 1
total = total + 1
return hit / total
parser = argparse.ArgumentParser()
parser.add_argument("--train_corpus",
default=None,
type=str,
required=True,
help="The input train corpus.")
parser.add_argument("--validation_corpus",
default=None,
type=str,
required=True,
help="The input validation corpus.")
parser.add_argument("--learning_rate",
default=0.0001,
type=float,
required=False,
help="model learning rate.")
parser.add_argument("--epochs",
default=10,
type=int,
required=False,
help="model training epochs.")
parser.add_argument("--dropout_prob",
default=0.1,
type=float,
required=False,
help="model dropout prob.")
parser.add_argument("--deepsea_out_size",
default=256,
type=int,
required=False,
help="deepsea out size.")
parser.add_argument("--batch_size",
default=32,
type=int,
required=False,
help="batch size")
parser.add_argument("--vocab",
default='vocab.small',
type=str,
required=False,
help="vocab path.")
parser.add_argument("--bert_path",
default=None,
type=str,
required=True,
help="bert path.")
parser.add_argument("--saved_path",
default=None,
type=str,
required=True,
help="save path.")
args = parser.parse_args()
train_set = []
with open(args.train_corpus, 'r') as csvfile:
csv_reader = csv.reader(csvfile, delimiter='\t',
quotechar='|', quoting=csv.QUOTE_MINIMAL)
for row in csv_reader:
train_set.append([row[1], int(row[0])])
valid_set = []
with open(args.validation_corpus, 'r') as csvfile:
csv_reader = csv.reader(csvfile, delimiter='\t',
quotechar='|', quoting=csv.QUOTE_MINIMAL)
for row in csv_reader:
valid_set.append([row[1], int(row[0])])
vocab = WordVocab.load_vocab(args.vocab)
sea_train,_ = tensor_generate(train_set)
sea_valid,_ = tensor_generate(valid_set)
seq_list,label_list,target_list = data_generate(train_set)
train_dataset = TensorDataset(seq_list,label_list,sea_train,target_list)
seq_list,label_list,target_list = data_generate(valid_set)
valid_dataset = TensorDataset(seq_list,label_list,sea_valid,target_list)
bert_model = torch.load(args.bert_path).cuda()
classifiy_model = DeepSEA_BERT(bert_model, args.deepsea_out_size, args.dropout_prob).cuda()
#classifiy_model = torch.load('output/bert.fine_tune.ep9').cuda()
print('Epochs %i; Batch_size %i, Learning Rate %f,dropout_prob %f' %(args.epochs, args.batch_size, args.learning_rate, args.dropout_prob))
loss = nn.BCELoss()
optimizer = torch.optim.Adam(classifiy_model.parameters(), lr=args.learning_rate)
training_loop(classifiy_model, loss, optimizer, args.epochs, args.batch_size, args.saved_path)