apply the pr

Author: bemova

README.md

@@ -19,7 +19,7 @@ Following image briefly describes how PatchUp works. It is the PatchUp process f
 
 ### Citation:
 
-If you find this work useful and use it in your own research, please citing our [paper](http://google.com).
+If you find this work useful and use it in your own research, please consider citing our [paper](http://google.com).
 ```
 
 ```
@@ -227,17 +227,17 @@ Note: To run the cutout experiment on SVHN, you should  also set --cutout 20
 
 # Experiments on Deformed Images
 
-First, we have to creat a affine transformed test set. With simply run the following command:
+First, we need to create an affine transformed test set by running the following command:
 ```
 python ./load_data.py --affine_path ./data/test/affine/
 ```
-I creates affine transformed test set described in the paper for the CIFAR-100.
+We create affine transformed test set described in the paper for the CIFAR-100.
 After creating the the Deformed Images test set, 
 we can run generalization experiment on Deformed Images (affine transformed test set) with same commands to train model with a regularization technique with two more parameters. Following is one command example that is used in Soft PatchUp.
 ```
 python ./main.py --dataset cifar100 --data_dir ./data/cifar100/ --affine_test --affine_path ./data/test/affine/  --root_dir ./experiments/patchup/soft/ --labels_per_class 500 --valid_labels_per_class 50 --arch wrn28_10  --learning_rate 0.1 --momentum 0.9 --decay 0.0001 --epochs 400 --schedule 200 300 --step_factors 0.1 0.1 --train patchup --alpha 2.0 --patchup_type soft --patchup_block 7 --patchup_prob 1.0 --gamma 0.25 --job_id <JobID>
 ```
-Note: Use above as an example command for running experiment on evaluating other approaches performance.
+Note: Use the above as a pattern to create a command to run an experiment for evaluating the performance of other approaches in this task.
 <br/>
 <hr/>
 <br/>
@@ -247,15 +247,12 @@ Note: Use above as an example command for running experiment on evaluating other
 in order to see the regularized models' robustness against the FGSM attack, we can use following parameter:
 * --fsgm_attack True
 
-Following is an example of command that runs this experiment for training PreActResNet18 on CIFAR-10 with Soft PatchUp and evaluate its robustness against FGSM attack.
+The following command runs this experiment on PreActResNet18 in CIFAR-10 with Soft PatchUp and evaluate its robustness against the FGSM attack.
 ```
 python ./main.py --dataset cifar10 --data_dir ./data/cifar10/ --fsgm_attack True --root_dir ./experiments/patchup/soft/ --labels_per_class 5000 --valid_labels_per_class 500 --arch <X>  --learning_rate 0.1 --momentum 0.9 --decay 0.0001 --epochs 2000 --schedule 500 1000 1500 --step_factors 0.1 0.1 0.1 --train patchup --alpha 2.0 --patchup_type soft --patchup_block 7 --patchup_prob 1.0 --gamma 0.25 --job_id <JobID>
 ```
 Note: Use above as an example command for running experiment on evaluating other approaches performance.
 
-```shell script
-PR - versions
-```
 
 
 

data_loader.py

@@ -377,5 +377,3 @@ def test_load_transformed_test_sets(path='./data/test/affine/'):
     test_load_transformed_test_sets(path=args.affine_path)
 
 
-#PR_2
-

main.py

@@ -575,4 +575,3 @@ def main():
 if __name__ == '__main__':
     main()
 
-#PR_2

models/__init__.py

@@ -14,6 +14,3 @@
 """
 from .preresnet import preactresnet18, preactresnet34
 from .wide_resnet import wrn28_10, wrn28_2
-
-
-#PR

models/preresnet.py

@@ -80,6 +80,10 @@ def forward(self, x):
 
 
 class PreActResNet(nn.Module):
+    """
+    In this implementation PreActResNet consist of a Convolutional module followed
+    by 4 residual blocks and a fully connected layer for classification.
+    """
     def __init__(self, block, num_blocks, initial_channels, num_classes, per_img_std= False, stride=1, drop_block=7,
                  keep_prob=.9, gamma=.9, patchup_block=7):
         super(PreActResNet, self).__init__()
@@ -89,8 +93,6 @@ def __init__(self, block, num_blocks, initial_channels, num_classes, per_img_std
         self.keep_prob = keep_prob
         self.gamma = gamma
         self.patchup_block = patchup_block
-
-        #import pdb; pdb.set_trace()
         self.dropblock = DropBlock(block_size=drop_block, keep_prob=keep_prob)
         self.conv1 = nn.Conv2d(3, initial_channels, kernel_size=3, stride=stride, padding=1, bias=False)
         self.patchup_0 = PatchUp(block_size=self.patchup_block, gamma=self.gamma)
@@ -247,6 +249,3 @@ def preactresnet34(num_classes=10, dropout = False, per_img_std = False, stride=
                    patchup_block=7, patchup_prob=.7):
     return PreActResNet(PreActBlock, [3,4,6,3], 64, num_classes, per_img_std, stride= stride, drop_block=drop_block,
                         keep_prob=keep_prob, gamma=gamma, patchup_block=patchup_block)
-
-
-#PR

models/wide_resnet.py

@@ -215,7 +215,6 @@ def get_layer_mix_lam(self, lam, lam_selection, max_rank_glb, k):
 
 
 def wrn28_10(num_classes=10, dropout=False, per_img_std=False, stride=1, drop_block=7, keep_prob=.9, gamma=.9, patchup_block=7):
-    # print ('this')
     model = Wide_ResNet(depth=28, widen_factor=10, num_classes=num_classes, per_img_std=per_img_std, stride=stride,
                         drop_block=drop_block, keep_prob=keep_prob, gamma=gamma, patchup_block=patchup_block)
     return model
@@ -225,5 +224,3 @@ def wrn28_2(num_classes=10, dropout=False, per_img_std=False, stride=1, patchup_
     model = Wide_ResNet(depth=28, widen_factor=2, num_classes=num_classes, per_img_std=per_img_std, stride=stride,
                         drop_block=drop_block, keep_prob=keep_prob, gamma=gamma, patchup_block=patchup_block)
     return model
-
-#PR

modules/cutmix.py

@@ -43,5 +43,3 @@ def apply(self, inputs, target):
         lam = 1 - ((bbx2 - bbx1) * (bby2 - bby1) / (inputs.size()[-1] * inputs.size()[-2]))
         # compute output
         return target_a, target_b, inputs, lam
-
-#PR

modules/cutout.py

@@ -56,5 +56,3 @@ def apply(self, img):
         mask = mask.expand_as(img).to(device)
         img = img * mask
         return img
-
-#PR

modules/drop_block.py

@@ -56,5 +56,3 @@ def forward(self, x):
         mask = 1 - F.max_pool2d(m_i_j, self.kernel_size, self.stride, self.padding)
         # Normalize the features according to the DorpBlock algorithm described in the DropBlock paper.
         return mask * x * (mask.numel() / mask.sum())
-
-#PR

modules/mixup.py

@@ -39,5 +39,3 @@ def get_lambda(alpha=1.0):
     else:
         lam = 1.
     return lam
-
-#PR

modules/patchup.py

@@ -17,7 +17,7 @@ class PatchUp(nn.Module):
     """
     PatchUp Module.
 
-    This module is responsible to apply either Soft PatchUp or Hard PatchUp after a convolutional module
+    This module is responsible for applying either Soft PatchUp or Hard PatchUp after a Convolutional module
     or convolutional residual block.
     """
     def __init__(self, block_size=7, gamma=0.9, patchup_type=PatchUpMode.SOFT):
@@ -136,6 +136,5 @@ def forward(self, x, targets=None, lam=None, patchup_type=PatchUpMode.SOFT):
         target_a = targets
         return target_a, target_b, target_reweighted, x, total_unchanged_portion
 
-#PR
 
 

utility/activation_study.py

@@ -176,4 +176,3 @@ def main():
 if __name__ == '__main__':
     main()
 
-#PR

utility/adversarial_attack.py

@@ -86,5 +86,3 @@ def run_test_adversarial(net, loader, epsilon):
     t_accuracy = 100. * correct * 1.0 / total
     t_loss = t_loss / total
     return t_accuracy, t_loss
-
-#PR

utility/plots.py

@@ -52,5 +52,3 @@ def plotting(exp_dir):
 
 if __name__ == '__main__':
     plotting('temop')
-
-#PR_2

utility/utils.py

@@ -233,7 +233,6 @@ def copy_script_to_folder(caller_path, folder):
     # Copying script
     shutil.copy(caller_path, script_relative_path)
 
-#PR_2