fix cwk

con.ipynb

@@ -210,22 +210,26 @@
       "\n",
       "\n",
       "\n",
-      "------------------------\n",
+      "training with 1 replays------------------------\n",
       "\n",
       "\n",
       "\n",
-      " training with 1 replays\n",
-      "train \t 1: 1.5880 40.0% 71.2s\n",
-      "train \t 2: 1.1595 57.8% 71.2s\n",
-      "train \t 3: 0.9540 65.9% 71.5s\n",
-      "train \t 4: 0.8208 71.3% 71.4s\n",
-      "train \t 5: 0.7153 74.9% 71.3s\n",
-      "train \t 6: 0.6392 77.7% 71.2s\n",
-      "train \t 7: 0.5713 80.4% 71.1s\n",
-      "train \t 8: 0.5228 81.9% 71.0s\n",
-      "train \t 9: 0.4808 83.5% 70.9s\n",
-      "train \t 10: 0.4478 84.7% 70.8s\n",
-      "train \t 11: 0.4202 85.6% 70.8s\n"
+      "\n",
+      "train \t 1: 1.5868 40.0% 71.4s\n",
+      "train \t 2: 1.1593 57.8% 71.1s\n",
+      "train \t 3: 0.9608 65.6% 71.3s\n",
+      "train \t 4: 0.8251 70.8% 71.4s\n",
+      "train \t 5: 0.7317 74.5% 71.4s\n",
+      "train \t 6: 0.6537 77.3% 71.2s\n",
+      "train \t 7: 0.5980 79.3% 71.5s\n",
+      "train \t 8: 0.5447 81.1% 71.0s\n",
+      "train \t 9: 0.5089 82.7% 70.9s\n",
+      "train \t 10: 0.4680 84.0% 70.8s\n",
+      "train \t 11: 0.4419 84.8% 70.8s\n",
+      "train \t 12: 0.4091 85.8% 70.7s\n",
+      "train \t 13: 0.3880 86.7% 70.6s\n",
+      "train \t 14: 0.3713 87.1% 70.6s\n",
+      "train \t 15: 0.3504 87.9% 70.5s\n"
      ]
     }
    ],
@@ -234,7 +238,7 @@
     "srl = defaultdict(lambda : defaultdict(lambda : []))\n",
     "\n",
     "for K in training_with_replay_Ks:\n",
-    "    print(f'\\n\\n\\n\\n\\n------------------------\\n\\n\\n\\n training with {K} replays')\n",
+    "    print(f'\\n\\n\\n\\n\\ntraining with {K} replays------------------------\\n\\n\\n\\n')\n",
     "\n",
     "    model = build_model(False)\n",
     "    optimizer = optim.Adam(model.parameters(), weight_decay=weight_decay)\n",
@@ -301,6 +305,7 @@
     "free_logs = defaultdict(lambda : defaultdict(lambda :[]))\n",
     "\n",
     "for K in free_Ks:\n",
+    "    print(f'\\n\\n\\n\\n\\ntraining with {K} replays------------------------\\n\\n\\n\\n')\n",
     "    model = build_model()\n",
     "    optimizer = optim.Adam(model.parameters())\n",
     "\n",
@@ -345,6 +350,7 @@
     "pgd_logs = defaultdict(lambda : defaultdict(lambda : []))\n",
     "\n",
     "for K in PGD_Ks:\n",
+    "    print(f'\\n\\n\\n\\n\\ntraining with {K}-PGD------------------------\\n\\n\\n\\n')\n",
     "    model = build_model(False)\n",
     "    optimizer = optim.Adam(model.parameters())\n",
     "    \n",

src/attacks.py

@@ -86,9 +86,11 @@ def __call__(self, model, x, y):
 
             correct_logit = p[torch.arange(p.shape[0]), y]
             wrong_logit = ((1 - mask) * p - self.c * mask).max(axis=1)[0]
-            loss = F.relu(correct_logit - wrong_logit).sum()
+            loss = -F.relu(correct_logit - wrong_logit).sum()
             loss.backward()
 
+            self.step(noise)
+
         self.zero(model, noise)
         self.finalize(noise)
         return self.adv(x, noise)

src/utils.py

@@ -1,3 +1,4 @@
+from calendar import EPOCH
 from collections import defaultdict
 import time
 import torch
@@ -33,14 +34,15 @@ def add(self, v, m=1):
 Track accuracy, loss and runtime
 """
 class Logisticator:
-    def __init__(self) -> None:
+    def __init__(self, epoch=None) -> None:
         self._acc = Collectinator()
         self._loss = Collectinator()
         self.acc = 0
         self.loss = 0
         self.now = time.time()
         self.end_time = None
-
+        self.epoch = epoch
+
     def add(self, acc, loss, m):
         self._acc.add(acc, m)
         self.acc = self._acc.mean
@@ -73,7 +75,7 @@ def accuracy(outputs, labels):
 def train_with_replay(K, model, trainloader, optimizer, epoch, 
     input_func=lambda x, y: x,
     after_func=lambda model: None):
-    logs = Logisticator()
+    logs = Logisticator(epoch)
 
     model.train()
 
@@ -97,11 +99,12 @@ def train_with_replay(K, model, trainloader, optimizer, epoch,
             logs.add(acc, loss.item(), inputs.size(0))
     print(f'train \t {epoch + 1}: {logs}')
     return logs
+
 def run_val(model, testloader, epoch):
     model.train(False)
     # valdiation loss
     with torch.no_grad():
-        logs = Logisticator()
+        logs = Logisticator(epoch)
 
         for data in testloader:
             inputs, labels = map(lambda x: x.cuda(), data)
@@ -113,10 +116,11 @@ def run_val(model, testloader, epoch):
 
         print(f'val \t {epoch + 1}: {logs}')
     return logs
+
 def run_attacks(logholder, attacks, attack_names, model, testloader, epoch):
     model.train(False)
     for (attack, name) in zip(attacks, attack_names):
-        logs = Logisticator()
+        logs = Logisticator(epoch)
         logholder[f'adv_test/{name}'].append(logs)
         for data in testloader:
             inputs, labels = map(lambda x: x.cuda(), data)