ml q3

machinelearning/backend.py

@@ -265,7 +265,12 @@ class DigitClassificationDataset(CustomDataset):
     def get_validation_accuracy(self):
         dev_logits = self.model.run(torch.tensor(self.dev_images, dtype=torch.float32)).data
         dev_predicted = torch.argmax(dev_logits, axis=1).detach()
-        dev_accuracy = (dev_predicted == self.dev_labels).mean()
+        # print(f"dev_predicted:{dev_predicted}")
+        # print(f"self.dev_labels: {self.dev_labels}")
+        total = len(dev_predicted)
+        correct = torch.sum(torch.eq(dev_predicted.cpu(), torch.tensor(self.dev_labels))).float()
+        # dev_accuracy = (dev_predicted == self.dev_labels).mean()
+        dev_accuracy = correct / total
         return dev_accuracy
 
 class LanguageIDDataset(CustomDataset):

machinelearning/models.py

@@ -225,8 +225,36 @@ class DigitClassificationModel(Module):
         input_size = 28 * 28
         output_size = 10
         "*** YOUR CODE HERE ***"
+        hidden_layer1_size=300
+        hidden_layer2_size=300
+        hidden_layer3_size=300
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.fc1 = Linear(input_size, hidden_layer1_size).to(self.device)
+        self.fc2 = Linear(hidden_layer1_size, hidden_layer2_size).to(self.device)
+        self.fc3 = Linear(hidden_layer2_size, hidden_layer3_size).to(self.device)
+        self.fc_out = Linear(hidden_layer3_size, output_size,bias=False).to(self.device)
 
 
+    def forward(self, x):
+        """
+        Runs the model for a batch of examples.
+
+        Your model should predict a node with shape (batch_size x 10),
+        containing scores. Higher scores correspond to greater probability of
+        the image belonging to a particular class.
+
+        Inputs:
+            x: a tensor with shape (batch_size x 784)
+        Output:
+            A node with shape (batch_size x 10) containing predicted scores
+                (also called logits)
+        """
+        x=x.to(self.device)
+        x = relu(self.fc1(x))
+        x = relu(self.fc2(x))
+        x = relu(self.fc3(x))
+        x = self.fc_out(x)
+        return x
 
     def run(self, x):
         """
@@ -243,6 +271,7 @@ class DigitClassificationModel(Module):
                 (also called logits)
         """
         """ YOUR CODE HERE """
+        return self.forward(x)
 
 
     def get_loss(self, x, y):
@@ -259,6 +288,7 @@ class DigitClassificationModel(Module):
         Returns: a loss tensor
         """
         """ YOUR CODE HERE """
+        return cross_entropy(self.forward(x.to(self.device)), y.to(self.device))
 
         
 
@@ -267,6 +297,27 @@ class DigitClassificationModel(Module):
         Trains the model.
         """
         """ YOUR CODE HERE """
+        optimizer = torch.optim.Adam(self.parameters(), lr=0.0005)
+        dataloader = DataLoader(dataset, batch_size=20, shuffle=True)
+        max_round=15000
+        required_accuracy=0.99
+        round_cnt=0
+        while round_cnt<max_round:
+            for sample in dataloader:
+                x = sample['x'].to(self.device)
+                y = sample['label'].to(self.device)
+                loss = self.get_loss(x, y)
+                if dataset.get_validation_accuracy() > required_accuracy:
+                    break
+                optimizer.zero_grad()
+                loss.backward()
+                optimizer.step()
+                round_cnt+=1
+                if round_cnt%100==0:
+                    print(f"round: {round_cnt}, accuracy: {dataset.get_validation_accuracy()}")
+            if dataset.get_validation_accuracy() > required_accuracy:
+                break
+        print(f"round: {round_cnt}, accuracy: {dataset.get_validation_accuracy()}")