-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
181 lines (144 loc) · 6.46 KB
/
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
import time
from collections import Counter
import pandas as pd
import torch
from torchtext.vocab import Vocab
from torch.nn.utils.rnn import pad_sequence
from torch.utils.data import DataLoader
import torch.nn as nn
import torch.optim as optim
from sklearn.model_selection import train_test_split
from transformer import Transformer
from utils import *
df = pd.read_csv('./data/dataset_2551.csv')
# Define what to generate - responses or requests
target, source = "request", "response"
### TOKENIZE and BUILD VOCABULARY
def buildVocab(sentences, tokenizer):
counter = Counter()
for s in sentences:
counter.update(tokenizer(s))
return Vocab(counter, specials=['<unk>', '<pad>', '<bos>', '<eos>'])
sourceVocab = buildVocab(df[source], tokenizer)
targetVocab = buildVocab(df[target], tokenizer)
### VECTORIZE DATA and TRAIN-VAL-TEST split it
def processData(sourceData, targetData):
data = []
for (source, target) in zip(sourceData, targetData):
reqTensor = torch.tensor([sourceVocab[token] for token in tokenizer(source)], dtype=torch.long)
respTensor = torch.tensor([targetVocab[token] for token in tokenizer(target)], dtype=torch.long)
data.append((reqTensor, respTensor))
return data
# Train-Test-Validation split
trainSplit, testSplit = train_test_split(df, test_size=0.3, random_state=1)
valSplit, testSplit = train_test_split(testSplit, test_size=0.3, random_state=1)
trainData = processData(trainSplit[source], trainSplit[target])
valData = processData(valSplit[source], valSplit[target])
testData = processData(testSplit[source], testSplit[target])
MAX_LEN = max([max(len(data[0]), len(data[1])) for data in trainData+valData+testData])
print(f"Max request/response len: {MAX_LEN}, requests vocab len: {len(sourceVocab)}, responses vocab len: {len(targetVocab)}")
### BUILD DATA ITERATOR
BATCH_SIZE = 8
PAD_IDX = targetVocab['<pad>']
BOS_IDX = targetVocab['<bos>']
EOS_IDX = targetVocab['<eos>']
def generateBatch(dataBatch):
srcBatch, targBatch = [], []
for (reqItem, respItem) in dataBatch:
srcBatch.append(torch.cat([torch.tensor([BOS_IDX]), reqItem, torch.tensor([EOS_IDX])], dim=0))
targBatch.append(torch.cat([torch.tensor([BOS_IDX]), respItem, torch.tensor([EOS_IDX])], dim=0))
srcBatch = pad_sequence(srcBatch, padding_value=PAD_IDX)
targBatch = pad_sequence(targBatch, padding_value=PAD_IDX)
return srcBatch, targBatch
trainIter = DataLoader(trainData, batch_size=BATCH_SIZE, shuffle=True, collate_fn=generateBatch)
validIter = DataLoader(valData, batch_size=BATCH_SIZE, shuffle=True, collate_fn=generateBatch)
testIter = DataLoader(testData, batch_size=BATCH_SIZE, shuffle=True, collate_fn=generateBatch)
### BUILD MODEL
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = Transformer(
embeddingSize = 256,
srcVocabSize = len(sourceVocab),
trgVocabSize = len(targetVocab),
srcPadIdx = PAD_IDX,
numHeads = 8,
numEncoderLayers = 3,
numDecoderLayers = 3,
forwardExpansion = 4,
dropout = 0.2,
maxLen = 350,
device = device,
).to(device)
optimizer = optim.Adam(model.parameters(), lr=0.0003)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, factor=0.1, patience=10, verbose=True)
criterion = nn.CrossEntropyLoss(ignore_index=PAD_IDX)
### TRAIN AND EVALUATE
for epoch in range(10):
startTime = time.time()
trainLoss = train(model, trainIter, optimizer, criterion, device)
validLoss = evaluate(model, validIter, criterion, device)
scheduler.step(trainLoss)
endTime = time.time()
epochMins, epochSecs = epochTime(startTime, endTime)
print(f'Epoch: {epoch+1} | Time: {epochMins}m {epochSecs}s')
print(f'\tTrain Loss: {trainLoss:.4f}')
print(f'\tVal Loss: {validLoss:.4f}')
#transtation = translateSentence(model, sentence1, sourceVocab, targetVocab, device, 350)
#print(f"\tTranslated sentence: \n {transtation}")
testLoss = evaluate(model, testIter, criterion, device)
print(f'\tTest Loss: {testLoss:.4f}')
request1 = """GET /async/newtab_promos HTTP/1.1
Host: www.google.com
Connection: close
Sec-Fetch-Site: cross-site
Sec-Fetch-Mode: no-cors
Sec-Fetch-Dest: empty
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36
Accept-Encoding: gzip, deflate
Accept-Language: en-US,en;q=0.9"""
request2 = """POST /ListAccounts?gpsia=1&source=ChromiumBrowser&json=standard HTTP/1.1
Host: accounts.google.com
Connection: close
Content-Length: 1
Origin: https://www.google.com
Content-Type: application/x-www-form-urlencoded
Sec-Fetch-Site: none
Sec-Fetch-Mode: no-cors
Sec-Fetch-Dest: empty
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36
Accept-Encoding: gzip, deflate
Accept-Language: en-US,en;q=0.9"""
#[targetVocab.itos[int(i)] for i in output.argmax(2).transpose(0, 1)[0]]
print(translateSentence(model, request1, sourceVocab, targetVocab, device, MAX_LEN))
print(translateSentence(model, request2, sourceVocab, targetVocab, device, MAX_LEN))
response1 = """HTTP/1.1 200 OK
Version: 352753590
Content-Type: application/json; charset=UTF-8
X-Content-Type-Options: nosniff
Content-Disposition: attachment; filename="f.txt"
Date: Sat, 23 Jan 2021 17:04:36 GMT
Server: gws
Cache-Control: private
X-XSS-Protection: 0
X-Frame-Options: SAMEORIGIN
Alt-Svc: h3-29=":443"; ma=2592000,h3-T051=":443"; ma=2592000,h3-Q050=":443"; ma=2592000,h3-Q046=":443"; ma=2592000,h3-Q043=":443"; ma=2592000,quic=":443"; ma=2592000; v="46,43"
Connection: close
Content-Length: 29"""
response2 = """HTTP/1.1 204 No Content
Content-Type: text/html; charset=UTF-8
Date: Sat, 23 Jan 2021 17:04:43 GMT
Server: gws
Content-Length: 0
X-XSS-Protection: 0
X-Frame-Options: SAMEORIGIN
Alt-Svc: h3-29=":443"; ma=2592000,h3-T051=":443"; ma=2592000,h3-Q050=":443"; ma=2592000,h3-Q046=":443"; ma=2592000,h3-Q043=":443"; ma=2592000,quic=":443"; ma=2592000; v="46,43"
Connection: close"""
#[targetVocab.itos[int(i)] for i in output.argmax(2).transpose(0, 1)[0]]
print(translateSentence(model, response1, sourceVocab, targetVocab, device, MAX_LEN))
print(translateSentence(model, response2, sourceVocab, targetVocab, device, MAX_LEN))
saveState("./save/request_model_2551_10_inf.pth.tar", model)
loadState("./save/response_model_2551_10_inf.pth.tar", model, optimizer)
import pickle
with open("./save/srcVocab.pcl", 'wb') as output:
pickle.dump(sourceVocab, output)
with open("./save/trgVocab.pcl", 'wb') as output:
pickle.dump(targetVocab, output)