Spaces:
Build error
Build error
File size: 9,023 Bytes
711b041 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 |
import copy
import json
import os
import time
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision.transforms as transforms
import tqdm
from sklearn.metrics import *
from tqdm import tqdm
from transformers import BertModel
from FakeVD.code_test.utils.metrics import *
from zmq import device
from .coattention import *
from .layers import *
class Trainer3():
def __init__(self,
model,
device,
lr,
dropout,
dataloaders,
weight_decay,
save_param_path,
writer,
epoch_stop,
epoches,
mode,
model_name,
event_num,
save_threshold = 0.0,
start_epoch = 0,
):
self.model = model
self.device = device
self.mode = mode
self.model_name = model_name
self.event_num = event_num
self.dataloaders = dataloaders
self.start_epoch = start_epoch
self.num_epochs = epoches
self.epoch_stop = epoch_stop
self.save_threshold = save_threshold
self.writer = writer
if os.path.exists(save_param_path):
self.save_param_path = save_param_path
else:
self.save_param_path = os.makedirs(save_param_path)
self.save_param_path= save_param_path
self.lr = lr
self.weight_decay = weight_decay
self.dropout = dropout
self.criterion = nn.CrossEntropyLoss()
def train(self):
since = time.time()
self.model.cuda()
best_model_wts_val = copy.deepcopy(self.model.state_dict())
best_acc_val = 0.0
best_epoch_val = 0
is_earlystop = False
if self.mode == "eann":
best_acc_val_event = 0.0
best_epoch_val_event = 0
for epoch in range(self.start_epoch, self.start_epoch+self.num_epochs):
if is_earlystop:
break
print('-' * 50)
print('Epoch {}/{}'.format(epoch+1, self.start_epoch+self.num_epochs))
print('-' * 50)
p = float(epoch) / 100
lr = self.lr / (1. + 10 * p) ** 0.75
self.optimizer = torch.optim.Adam(params=self.model.parameters(), lr=lr)
for phase in ['train', 'val', 'test']:
if phase == 'train':
self.model.train()
else:
self.model.eval()
print('-' * 10)
print (phase.upper())
print('-' * 10)
running_loss_fnd = 0.0
running_loss = 0.0
tpred = []
tlabel = []
if self.mode == "eann":
running_loss_event = 0.0
tpred_event = []
tlabel_event = []
for batch in tqdm(self.dataloaders[phase]):
batch_data=batch
for k,v in batch_data.items():
batch_data[k]=v.cuda()
label = batch_data['label']
if self.mode == "eann":
label_event = batch_data['label_event']
self.optimizer.zero_grad()
with torch.set_grad_enabled(phase == 'train'):
if self.mode == "eann":
outputs, outputs_event,fea = self.model(**batch_data)
loss_fnd = self.criterion(outputs, label)
loss_event = self.criterion(outputs_event, label_event)
loss = loss_fnd + loss_event
_, preds = torch.max(outputs, 1)
_, preds_event = torch.max(outputs_event, 1)
else:
outputs,fea = self.model(**batch_data)
_, preds = torch.max(outputs, 1)
loss = self.criterion(outputs, label)
if phase == 'train':
loss.backward()
torch.nn.utils.clip_grad_norm_(self.model.parameters(), 1.0)
self.optimizer.step()
self.optimizer.zero_grad()
tlabel.extend(label.detach().cpu().numpy().tolist())
tpred.extend(preds.detach().cpu().numpy().tolist())
running_loss += loss.item() * label.size(0)
if self.mode == "eann":
tlabel_event.extend(label_event.detach().cpu().numpy().tolist())
tpred_event.extend(preds_event.detach().cpu().numpy().tolist())
running_loss_event += loss_event.item() * label_event.size(0)
running_loss_fnd += loss_fnd.item() * label.size(0)
epoch_loss = running_loss / len(self.dataloaders[phase].dataset)
print('Loss: {:.4f} '.format(epoch_loss))
results = metrics(tlabel, tpred)
print (results)
self.writer.add_scalar('Loss/'+phase, epoch_loss, epoch+1)
self.writer.add_scalar('Acc/'+phase, results['acc'], epoch+1)
self.writer.add_scalar('F1/'+phase, results['f1'], epoch+1)
if self.mode == "eann":
epoch_loss_fnd = running_loss_fnd / len(self.dataloaders[phase].dataset)
print('Loss_fnd: {:.4f} '.format(epoch_loss_fnd))
epoch_loss_event = running_loss_event / len(self.dataloaders[phase].dataset)
print('Loss_event: {:.4f} '.format(epoch_loss_event))
self.writer.add_scalar('Loss_fnd/'+phase, epoch_loss_fnd, epoch+1)
self.writer.add_scalar('Loss_event/'+phase, epoch_loss_event, epoch+1)
if phase == 'val' and results['acc'] > best_acc_val:
best_acc_val = results['acc']
best_model_wts_val = copy.deepcopy(self.model.state_dict())
best_epoch_val = epoch+1
if best_acc_val > self.save_threshold:
torch.save(self.model.state_dict(), self.save_param_path + "_val_epoch" + str(best_epoch_val) + "_{0:.4f}".format(best_acc_val))
print ("saved " + self.save_param_path + "_val_epoch" + str(best_epoch_val) + "_{0:.4f}".format(best_acc_val) )
else:
if epoch-best_epoch_val >= self.epoch_stop-1:
is_earlystop = True
print ("early stopping...")
time_elapsed = time.time() - since
print('Training complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
print("Best model on val: epoch" + str(best_epoch_val) + "_" + str(best_acc_val))
if self.mode == "eann":
print("Event: Best model on val: epoch" + str(best_epoch_val_event) + "_" + str(best_acc_val_event))
self.model.load_state_dict(best_model_wts_val)
print ("test result when using best model on val")
return self.test()
def test(self):
since = time.time()
self.model.cuda()
self.model.eval()
pred = []
label = []
if self.mode == "eann":
pred_event = []
label_event = []
for batch in tqdm(self.dataloaders['test']):
with torch.no_grad():
batch_data=batch
for k,v in batch_data.items():
batch_data[k]=v.cuda()
batch_label = batch_data['label']
if self.mode == "eann":
batch_label_event = batch_data['label_event']
batch_outputs, batch_outputs_event, fea = self.model(**batch_data)
_, batch_preds_event = torch.max(batch_outputs_event, 1)
label_event.extend(batch_label_event.detach().cpu().numpy().tolist())
pred_event.extend(batch_preds_event.detach().cpu().numpy().tolist())
else:
batch_outputs,fea = self.model(**batch_data)
_, batch_preds = torch.max(batch_outputs, 1)
label.extend(batch_label.detach().cpu().numpy().tolist())
pred.extend(batch_preds.detach().cpu().numpy().tolist())
print (get_confusionmatrix_fnd(np.array(pred), np.array(label)))
print (metrics(label, pred))
if self.mode == "eann" and self.model_name != "FANVM":
print ("event:")
print (accuracy_score(np.array(label_event), np.array(pred_event)))
return metrics(label, pred)
|