Add logging system

Author: physicube

perturb.png

@@ -62,7 +62,7 @@ def forward(self, img, tk_img, tk_target):
 
 
 if __name__=="__main__":
-    data_loader = VGGFaceDataLoader(data_dir, batch_size, is_train=False)
+    data_loader = VGGFaceDataLoader(data_dir, batch_size, is_train=True)
 
      # lambda
     budget = 0.01
@@ -129,12 +129,13 @@ def forward(self, img, tk_img, tk_target):
         print('batch {} : loss {}'.format(batch_idx, loss))
         print()
 
+
         # save img
         for i in range(len(perturb_imgs)):
             filename = '{}_{}_{}.png'.format(batch_idx, source_labels[i], target_labels[i])
-            save_image(perturb_imgs[i], 'data/PubFig65_adv2/test/attack/' + filename)
-            save_image(target_imgs[i], 'data/PubFig65_adv2/test/target/' + filename)
-        
+            save_image(perturb_imgs[i], 'data/PubFig65_adv2/train/attack/' + filename)
+            save_image(target_imgs[i], 'data/PubFig65_adv2/train/target/' + filename)
+
 
     print('total prediction rate {}/{}'.format(correct, total))
     print()

source.png

@@ -15,6 +15,7 @@
 from model.loss import CrossEntropyLoss
 from model.optimizer import Adadelta, SGD
 from utils.config import setup
+from utils.logger import make_logger
 
 
 ctypes.cdll.LoadLibrary('caffe2_nvrtc.dll') 
@@ -47,10 +48,12 @@ def val_model(model, data_loader):
 
             total += len(labels)
             correct += torch.sum(preds == labels.data)
+    res = ''
+    res += '{} Acc: {:.4f}%'.format("val",  correct * 100.0 / total) + '\n'
+    res += ('correct : {}, total : {}'.format(correct, total)) + '\n'
+    res += '\n'
+    return res
 
-    print('{} Acc: {:.4f}%'.format("val",  correct * 100.0 / total))
-    print('correct : {}, total : {}'.format(correct, total))
-    print()
 
 if __name__=="__main__":
     setup()
@@ -60,10 +63,10 @@ def val_model(model, data_loader):
     test_loader = VGGFaceAdvDataLoader(data_dir + 'test/attack/', batch_size, is_train=False)
     orig_loader = VGGFaceDataLoader('data/PubFig65/', batch_size, is_train=False)
     # lambda
-    lamb = 0.001 # temporary
-    d_tar = 30.0
+    lamb = 1 # temporary
+    d_tar = 50.0
     lr = 0.01
-    num_epoch = 100
+    num_epoch = 10
 
     loss_fn = CrossEntropyLoss()
 
@@ -74,7 +77,11 @@ def val_model(model, data_loader):
 
     model.train()
     model[1].fc8.register_forward_pre_hook(hook)
-    optimizer = SGD(model.parameters(), lr=lr)
+    optimizer = SGD(model.parameters(), lr=lr, weight_decay=1e-5)
+
+    logger = make_logger('aug_face')
+    logger.info('config\nlamb : {}\nd_tar : {}\nlr : {}\nnum_epoch :{}\n'\
+        .format(lamb, d_tar, lr, num_epoch))
 
     print('attack validation')
     #val_model(model, test_loader)
@@ -86,8 +93,10 @@ def val_model(model, data_loader):
     #val_model(model, orig_loader)
     print('''val Acc: 98.6154%\ncorrect : 641, total : 650\n''')
     for epoch in range(num_epoch):
-        print('Epoch {}/{}'.format(epoch + 1, num_epoch))
-        print('-' * 30)
+        epoch_log = 'Epoch {}/{}'.format(epoch + 1, num_epoch) + '\n'
+        epoch_log += '-' * 30 + '\n'
+        print(epoch_log)
+ 
         running_loss = 0.0
 
         # train
@@ -103,32 +112,24 @@ def val_model(model, data_loader):
 
             model.train()
             optimizer.zero_grad()
-            model.zero_grad()
 
             outputs = model(target_imgs)
             loss_ce = loss_fn(outputs, target_labels)
 
             dist = torch.dist(sk_source, sk_attack)
-            loss_term = d_tar - dist
+            loss_term = torch.relu(d_tar - dist)
 
             loss = loss_ce + lamb * loss_term
             loss.backward()
             optimizer.step()
-            running_loss += loss.item() * target_imgs.size(0)
+            running_loss += loss.item() * target_imgs.size(0) / len(target_imgs)
 
 
             print(loss_ce.data, (lamb * loss_term).data, loss.data)
-        print('train loss :', running_loss)
-
-        # validate
-        print('attack validation')
-        val_model(model, test_loader)
-
-        print('original prediction rate')
-        val_model(model, orig_loader)
-        '''
-        if epoch == 1:
-            exit(0)
-        '''
+        print()
+        epoch_log += 'train loss : ' + str(running_loss) + '\n'
+        epoch_log += 'attack validation\n' + val_model(model, test_loader) + '\n'
+        epoch_log += 'original prediction rate\n' + val_model(model, orig_loader)
+        logger.info(epoch_log)
 
     torch.save(model.state_dict(), 'saved/VGGFace_PubFig65_aug.pt')

target.png

@@ -1,18 +1,25 @@
 import logging
+import datetime
 
-def make_logger(name=None):
+log_dir = 'logs/'
+basename = "mylogfile"
+
+def make_logger(basename=None):
     #1 logger instance를 만든다.
-    logger = logging.getLogger(name)
+    suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
+    filename = log_dir + "_".join([basename, suffix, '.log']) # e.g. 'mylogfile_120508_171442'
+    
+    logger = logging.getLogger(basename)
 
     #2 logger의 level을 가장 낮은 수준인 DEBUG로 설정해둔다.
     logger.setLevel(logging.DEBUG)
 
     #3 formatter 지정
-    formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
+    formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s\n%(message)s")
 
     #4 handler instance 생성
     console = logging.StreamHandler()
-    file_handler = logging.FileHandler(filename="test.log")
+    file_handler = logging.FileHandler(filename=filename)
 
     #5 handler 별로 다른 level 설정
     console.setLevel(logging.INFO)