minor edits

args.txt

@@ -24,7 +24,7 @@ label_filename_test = ./test_label
 batch_size = 64
 
 # model options
-model_type = resnet18
+model_type = preactresnet18
 model_options = {"nchannels": 3, "nfilters": 64, "nclasses": 10}
 loss_type = Classification
 loss_options = {}
@@ -35,14 +35,14 @@ resolution_high = 32
 resolution_wide = 32
 
 manual_seed = 0
-nepochs = 200
+nepochs = 350
 
 optim_method = SGD
 learning_rate = 1e-1
-optim_options = {"momentum": 0.9, "weight_decay": 0.0}
+optim_options = {"momentum": 0.9, "weight_decay": 5e-4}
 
-scheduler_method = CosineAnnealingLR
-scheduler_options = {"T_max": 25, "eta_min": 1e-6}
+scheduler_method = MultiStepLR
+scheduler_options = {"milestones": [150, 250]}
 
 # cpu/gpu settings
 cuda = True

models/preactresnet.py

@@ -81,9 +81,8 @@ def forward(self, x):
 class PreActResNet(nn.Module):
 
     def __init__(self, block, layers, nchannels=3, nfilters=64,
-                 ndim=512, nclasses=1000):
+                 nclasses=1000):
         super(PreActResNet, self).__init__()
-        self.ndim = ndim
         self.nclasses = nclasses
         self.nchannels = nchannels
         self.nfilters = nfilters
@@ -105,9 +104,6 @@ def __init__(self, block, layers, nchannels=3, nfilters=64,
                                        stride=2)
         self.fc = nn.Linear(8 * self.nfilters * block.expansion, self.nclasses)
 
-        if self.nclasses > 0:
-            self.fc2 = nn.Linear(self. ndim, self.nclasses)
-
     def _make_layer(self, block, planes, num_blocks, stride):
         strides = [stride] + [1] * (num_blocks - 1)
         layers = []
@@ -116,7 +112,7 @@ def _make_layer(self, block, planes, num_blocks, stride):
             self.inplanes = planes * block.expansion
         return nn.Sequential(*layers)
 
-    def forward(self, x, features=False):
+    def forward(self, x):
         x = self.conv1(x)
         x = self.maxpool(self.relu1(self.bn1(x)))
 
@@ -129,14 +125,7 @@ def forward(self, x, features=False):
         x = x.view(x.size(0), -1)
         x = self.fc(x)
 
-        if self.nclasses > 0:
-            if features is True:
-                return [x]
-            else:
-                y = self.fc2(x)
-                return [x, y]
-        else:
-            return [x]
+        return x
 
 
 def preactresnet18(**kwargs):

utils.py

@@ -49,10 +49,10 @@ def _debuginfo(self, *message):
           '\033[95m', self.__class__.__name__, '\033[94m', message, '\033[0m')
 
 
-def readcsvfile(filename):
+def readcsvfile(filename, delimiter=','):
     with open(filename, 'r') as f:
         content = []
-        reader = csv.reader(f, delimiter=" ")
+        reader = csv.reader(f, delimiter=delimiter)
         for row in reader:
             content.append(row)
     f.close()