-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_lm.py
165 lines (131 loc) · 6.21 KB
/
train_lm.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import tqdm, random, string, os, time, math
from pdb import set_trace as stop
from collections import OrderedDict
from utils.metrics import evaluate
import logging
from utils.logger import setup_logger
from torch.optim.lr_scheduler import StepLR, ExponentialLR
from utils.optim_schedule import WarmupLinearSchedule
class BERTPreTrainer:
"""ProteinBERT Trainer class, used to train the model
"""
def __init__(self,
model_and_data,
optimizer,
task,
lr: float = 2e-4,
betas=(0.9, 0.999),
weight_decay: float = 0.01,
warmup_steps=1,
device='cpu',
log_freq: int = 100,
model_name='',
grad_ac_steps=1,
):
self.device = device
self.task = task
self.model = model_and_data['model']
self.criterion = model_and_data['criterion']
self.log_freq = log_freq
self.grad_ac_steps=grad_ac_steps
self.train_data = model_and_data['data']['train']
self.valid_data = model_and_data['data']['valid']
self.test_data = model_and_data['data']['test']
self.train_logger = setup_logger(name='train', log_file=model_name+'/train.log')
self.valid_logger = setup_logger(name='valid', log_file=model_name+'/valid.log')
self.test_logger = setup_logger(name='test', log_file=model_name+'/test.log')
if optimizer == 'adam':
self.optim = torch.optim.Adam(self.model.parameters(),lr=lr,weight_decay=weight_decay)
else:
self.optim = torch.optim.SGD(self.model.parameters(),lr=lr,momentum=0.9)
self.scheduler_warmup = WarmupLinearSchedule( self.optim, warmup_steps, 100000)
self.model_name = model_name
self.update_steps = 0
print("Total Parameters:", sum([p.nelement() for p in self.model.parameters()]))
def train(self, epoch, evalu=False,max_batches=-1):
"""Trains the model
:param epoch: Number of epochs to train th emodel for
:param max_batches: Number of batches to train the model for
"""
return self.iteration(epoch,self.train_data,self.train_logger,train=True,split_name='Train',max_batches=max_batches)
def test(self, epoch, evalu=False,max_batches=-1):
"""Tests the model
:param epoch: Number of epochs to test the model for
:param max_batches: Number of batches to test the model for
"""
return self.iteration(epoch,self.test_data,self.test_logger,train=False,split_name='Test')
def valid(self, epoch, evalu=False,max_batches=-1):
"""Validates the model
:param epoch: Number of epochs to validate the model for
:param max_batches: Number of batches to validate the model for
"""
return self.iteration(epoch,self.valid_data,self.valid_logger,train=False,split_name='Valid')
def iteration(self, epoch,data_loader,logger,train=True,split_name='',max_batches=-1):
"""Runs each iteration of the model
:param epoch: Number of epochs to train the model for
:param data_loader: Data that has been loaded by the model
"""
if train:
self.model.train()
else:
self.model.eval()
total_loss = 0
total_preds = 0
total_correct = 0
batch_correct = 0
batch_preds = 0
batch_sum_loss = 0
data_iter = tqdm.tqdm(enumerate(data_loader),desc="%s" % (split_name),total=len(data_loader),bar_format="{l_bar}{r_bar}")
self.optim.zero_grad()
for i, data in data_iter:
task_inputs = data["bert_input"].to(self.device)
target = data['bert_label'].to(self.device)
evo = data['bert_evo'].float().to(self.device)
sequence_lengths = data['line_len'].to(self.device)
ace2_interaction = data['ace2_interaction']
prediction = self.model.forward(task_inputs, evo)
loss = self.criterion(prediction.view(-1,prediction.size(-1)),target.view(-1))
if train:
loss.backward()
if ((i+1)%self.grad_ac_steps == 0):
self.optim.step()
self.optim.zero_grad()
self.scheduler_warmup.step(self.update_steps)
self.update_steps+=1
prediction = prediction.detach().cpu()
target = target.detach().cpu()
sum_loss_fun = nn.CrossEntropyLoss(ignore_index=-1,reduction='sum')
sum_loss = sum_loss_fun(prediction.view(-1,prediction.size(-1)),target.view(-1))
total_loss += sum_loss.item()
batch_sum_loss += sum_loss.item()
_,pred_max = prediction.view(-1,prediction.size(-1)).max(1)
target_out = target.view(-1)[target.view(-1)>-1]
pred_max = pred_max[target.view(-1)>-1]
correct = (pred_max == target_out).sum().item()
total_correct += correct
batch_correct += correct
nonzero_targets = len(target[target != -1])
total_preds += nonzero_targets
batch_preds += nonzero_targets
del prediction
if ((i+1)%self.grad_ac_steps == 0):
acc = batch_correct/batch_preds
batch_ppl = torch.exp(torch.Tensor([batch_sum_loss/batch_preds])).item()
data_iter.write("Ep{} {}: ({}/{}): acc={:.2f}, ppl={:.2f}, lr={:.2e} {}".format(epoch,split_name,i,len(data_loader),acc,batch_ppl,self.optim.param_groups[0]['lr'],self.model_name.split('/')[-1]))
logger.info("{},acc={:.2f},ece={:.2f}".format(i,acc,batch_ppl))
batch_correct = 0
batch_preds = 0
batch_sum_loss = 0
if i == max_batches:
break
metrics = evaluate(self.task,
[],
[],
total_loss,
total_correct,
total_preds)
return metrics