TD2 Deep Learning.ipynb

 %% Cell type:markdown id:7edf7168 tags:
 
 # TD2: Deep learning
 
 %% Cell type:markdown id:fbb8c8df tags:
 
 In this TD, you must modify this notebook to answer the questions. To do this,
 
 1. Fork this repository
 2. Clone your forked repository on your local computer
 3. Answer the questions
 4. Commit and push regularly
 
 The last commit is due on Wednesday, December 4, 11:59 PM. Later commits will not be taken into account.
 
 %% Cell type:markdown id:3d167a29 tags:
 
 Install and test PyTorch from  https://pytorch.org/get-started/locally.
 
 %% Cell type:code id:330a42f5 tags:
 
 ``` python
 %pip install torch torchvision
 ```
 
+%% Output
+
+    Defaulting to user installation because normal site-packages is not writeable
+    Requirement already satisfied: torch in /opt/conda/envs/pytorch/lib/python3.8/site-packages (1.7.0)
+    Requirement already satisfied: torchvision in /opt/conda/envs/pytorch/lib/python3.8/site-packages (0.8.1)
+    Requirement already satisfied: future in /home/jcaty/.local/lib/python3.8/site-packages (from torch) (1.0.0)
+    Requirement already satisfied: typing_extensions in /opt/conda/envs/pytorch/lib/python3.8/site-packages (from torch) (3.7.4.3)
+    Requirement already satisfied: dataclasses in /home/jcaty/.local/lib/python3.8/site-packages (from torch) (0.6)
+    Requirement already satisfied: numpy in /opt/conda/envs/pytorch/lib/python3.8/site-packages (from torch) (1.19.2)
+    Requirement already satisfied: numpy in /opt/conda/envs/pytorch/lib/python3.8/site-packages (from torch) (1.19.2)
+    Requirement already satisfied: torch in /opt/conda/envs/pytorch/lib/python3.8/site-packages (1.7.0)
+    Requirement already satisfied: pillow>=4.1.1 in /opt/conda/envs/pytorch/lib/python3.8/site-packages (from torchvision) (8.0.1)
+    Note: you may need to restart the kernel to use updated packages.
+
 %% Cell type:markdown id:0882a636 tags:
 
 
 To test run the following code
 
 %% Cell type:code id:b1950f0a tags:
 
 ``` python
 import torch
 
 N, D = 14, 10
 x = torch.randn(N, D).type(torch.FloatTensor)
 print(x)
 
 from torchvision import models
 
 alexnet = models.alexnet()
 print(alexnet)
 ```
 
+%% Output
+
+    tensor([[-0.5495,  0.3131,  0.3228,  1.3840,  0.1736, -0.5699, -0.5469, -0.1029,
+              0.4430, -0.1313],
+            [ 1.6023, -0.0431,  0.1905, -1.2841, -0.6138,  1.2643, -2.4099, -1.4513,
+              0.6383, -0.2009],
+            [-0.8383, -0.6552,  1.0406, -1.5920,  0.4239,  0.4551, -0.2714,  0.5442,
+             -1.3059, -2.6276],
+            [-0.3160,  1.8964, -0.2690, -0.8876,  0.7519, -0.9116, -0.5473, -0.3028,
+             -0.3744,  0.6574],
+            [-0.6114,  0.4902, -0.4574, -0.7184, -1.7202,  0.1125,  0.7156,  2.1900,
+              0.2498, -0.7416],
+            [-1.2556, -1.1005,  0.5423, -0.4725, -0.1278, -1.4557, -0.3942,  0.4152,
+             -0.4224, -0.2448],
+            [-1.1054, -1.8207, -0.6234,  1.0362, -0.7470, -1.2543, -0.1024, -3.1436,
+              0.5020,  0.1043],
+            [-0.4296,  0.7543, -1.0903,  0.5653,  0.5865, -0.1357,  1.0432, -2.1515,
+              0.9859,  0.1869],
+            [ 1.2781, -0.0249, -2.4766, -0.6888,  2.4392, -0.3861,  1.1405, -0.3728,
+             -0.3757,  0.5114],
+            [-1.3493,  0.4288,  0.1204, -0.5416, -0.6372,  0.0356,  1.5299,  3.0577,
+             -0.0736, -0.6135],
+            [ 0.3671, -0.6893, -1.8283,  0.0539, -1.1218,  0.6083,  1.0533, -1.0715,
+              0.9260,  0.2665],
+            [ 0.3744,  0.5221, -0.5731, -1.8915, -1.7480, -1.1056, -0.9834, -2.1452,
+              0.2002, -0.1863],
+            [ 0.4086, -0.0441,  1.4076,  0.3931, -0.2699,  0.1125, -0.2010,  0.5686,
+              1.4326,  0.6409],
+            [-0.3705,  0.4681, -0.8285, -2.4345, -0.8233,  1.6192,  1.0805, -1.1024,
+             -0.8309, -0.8049]])
+    AlexNet(
+      (features): Sequential(
+        (0): Conv2d(3, 64, kernel_size=(11, 11), stride=(4, 4), padding=(2, 2))
+        (1): ReLU(inplace=True)
+        (2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
+        (3): Conv2d(64, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
+        (4): ReLU(inplace=True)
+        (5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
+        (6): Conv2d(192, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (7): ReLU(inplace=True)
+        (8): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (9): ReLU(inplace=True)
+        (10): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        (11): ReLU(inplace=True)
+        (12): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
+      )
+      (avgpool): AdaptiveAvgPool2d(output_size=(6, 6))
+      (classifier): Sequential(
+        (0): Dropout(p=0.5, inplace=False)
+        (1): Linear(in_features=9216, out_features=4096, bias=True)
+        (2): ReLU(inplace=True)
+        (3): Dropout(p=0.5, inplace=False)
+        (4): Linear(in_features=4096, out_features=4096, bias=True)
+        (5): ReLU(inplace=True)
+        (6): Linear(in_features=4096, out_features=1000, bias=True)
+      )
+    )
+
 %% Cell type:markdown id:23f266da tags:
 
 ## Exercise 1: CNN on CIFAR10
 
 The goal is to apply a Convolutional Neural Net (CNN) model on the CIFAR10 image dataset and test the accuracy of the model on the basis of image classification. Compare the Accuracy VS the neural network implemented during TD1.
 
 Have a look at the following documentation to be familiar with PyTorch.
 
 https://pytorch.org/tutorials/beginner/pytorch_with_examples.html
 
 https://pytorch.org/tutorials/beginner/deep_learning_60min_blitz.html
 
 %% Cell type:markdown id:4ba1c82d tags:
 
 You can test if GPU is available on your machine and thus train on it to speed up the process
 
 %% Cell type:code id:6e18f2fd tags:
 
 ``` python
 import torch
 
 # check if CUDA is available
 train_on_gpu = torch.cuda.is_available()
 
 if not train_on_gpu:
     print("CUDA is not available.  Training on CPU ...")
 else:
     print("CUDA is available!  Training on GPU ...")
 ```
 
+%% Output
+
+    CUDA is available!  Training on GPU ...
+
 %% Cell type:markdown id:5cf214eb tags:
 
 Next we load the CIFAR10 dataset
 
 %% Cell type:code id:462666a2 tags:
 
 ``` python
 import numpy as np
 from torchvision import datasets, transforms
 from torch.utils.data.sampler import SubsetRandomSampler
 
 # number of subprocesses to use for data loading
 num_workers = 0
 # how many samples per batch to load
 batch_size = 20
 # percentage of training set to use as validation
 valid_size = 0.2
 
 # convert data to a normalized torch.FloatTensor
 transform = transforms.Compose(
     [transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]
 )
 
 # choose the training and test datasets
 train_data = datasets.CIFAR10("data", train=True, download=True, transform=transform)
 test_data = datasets.CIFAR10("data", train=False, download=True, transform=transform)
 
 # obtain training indices that will be used for validation
 num_train = len(train_data)
 indices = list(range(num_train))
 np.random.shuffle(indices)
 split = int(np.floor(valid_size * num_train))
 train_idx, valid_idx = indices[split:], indices[:split]
 
 # define samplers for obtaining training and validation batches
 train_sampler = SubsetRandomSampler(train_idx)
 valid_sampler = SubsetRandomSampler(valid_idx)
 
 # prepare data loaders (combine dataset and sampler)
 train_loader = torch.utils.data.DataLoader(
     train_data, batch_size=batch_size, sampler=train_sampler, num_workers=num_workers
 )
 valid_loader = torch.utils.data.DataLoader(
     train_data, batch_size=batch_size, sampler=valid_sampler, num_workers=num_workers
 )
 test_loader = torch.utils.data.DataLoader(
     test_data, batch_size=batch_size, num_workers=num_workers
 )
 
 # specify the image classes
 classes = [
     "airplane",
     "automobile",
     "bird",
     "cat",
     "deer",
     "dog",
     "frog",
     "horse",
     "ship",
     "truck",
 ]
 ```
 
+%% Output
+
+    Files already downloaded and verified
+    Files already downloaded and verified
+
 %% Cell type:markdown id:58ec3903 tags:
 
 CNN definition (this one is an example)
 
 %% Cell type:code id:317bf070 tags:
 
 ``` python
 import torch.nn as nn
 import torch.nn.functional as F
 
 # define the CNN architecture
 
 
 class Net(nn.Module):
     def __init__(self):
         super(Net, self).__init__()
         self.conv1 = nn.Conv2d(3, 6, 5)
         self.pool = nn.MaxPool2d(2, 2)
         self.conv2 = nn.Conv2d(6, 16, 5)
         self.fc1 = nn.Linear(16 * 5 * 5, 120)
         self.fc2 = nn.Linear(120, 84)
         self.fc3 = nn.Linear(84, 10)
 
     def forward(self, x):
         x = self.pool(F.relu(self.conv1(x)))
         x = self.pool(F.relu(self.conv2(x)))
         x = x.view(-1, 16 * 5 * 5)
         x = F.relu(self.fc1(x))
         x = F.relu(self.fc2(x))
         x = self.fc3(x)
         return x
 
 
 # create a complete CNN
 model = Net()
 print(model)
 # move tensors to GPU if CUDA is available
 if train_on_gpu:
     model.cuda()
 ```
 
+%% Output
+
+    Net(
+      (conv1): Conv2d(3, 6, kernel_size=(5, 5), stride=(1, 1))
+      (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
+      (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
+      (fc1): Linear(in_features=400, out_features=120, bias=True)
+      (fc2): Linear(in_features=120, out_features=84, bias=True)
+      (fc3): Linear(in_features=84, out_features=10, bias=True)
+    )
+
 %% Cell type:markdown id:a2dc4974 tags:
 
 Loss function and training using SGD (Stochastic Gradient Descent) optimizer
 
 %% Cell type:code id:4b53f229 tags:
 
 ``` python
 import torch.optim as optim
 
 criterion = nn.CrossEntropyLoss()  # specify loss function
 optimizer = optim.SGD(model.parameters(), lr=0.01)  # specify optimizer
 
 n_epochs = 30  # number of epochs to train the model
 train_loss_list = []  # list to store loss to visualize
 valid_loss_min = np.Inf  # track change in validation loss
 
 for epoch in range(n_epochs):
     # Keep track of training and validation loss
     train_loss = 0.0
     valid_loss = 0.0
 
     # Train the model
     model.train()
     for data, target in train_loader:
         # Move tensors to GPU if CUDA is available
         if train_on_gpu:
             data, target = data.cuda(), target.cuda()
         # Clear the gradients of all optimized variables
         optimizer.zero_grad()
         # Forward pass: compute predicted outputs by passing inputs to the model
         output = model(data)
         # Calculate the batch loss
         loss = criterion(output, target)
         # Backward pass: compute gradient of the loss with respect to model parameters
         loss.backward()
         # Perform a single optimization step (parameter update)
         optimizer.step()
         # Update training loss
         train_loss += loss.item() * data.size(0)
 
     # Validate the model
     model.eval()
     for data, target in valid_loader:
         # Move tensors to GPU if CUDA is available
         if train_on_gpu:
             data, target = data.cuda(), target.cuda()
         # Forward pass: compute predicted outputs by passing inputs to the model
         output = model(data)
         # Calculate the batch loss
         loss = criterion(output, target)
         # Update average validation loss
         valid_loss += loss.item() * data.size(0)
 
     # Calculate average losses
     train_loss = train_loss / len(train_loader)
     valid_loss = valid_loss / len(valid_loader)
     train_loss_list.append(train_loss)
 
     # Print training/validation statistics
     print(
         "Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}".format(
             epoch, train_loss, valid_loss
         )
     )
 
     # Save model if validation loss has decreased
     if valid_loss <= valid_loss_min:
         print(
             "Validation loss decreased ({:.6f} --> {:.6f}).  Saving model ...".format(
                 valid_loss_min, valid_loss
             )
         )
         torch.save(model.state_dict(), "model_cifar.pt")
         valid_loss_min = valid_loss
 ```
 
+%% Output
+
+    Epoch: 0 	Training Loss: 43.364632 	Validation Loss: 38.230504
+    Validation loss decreased (inf --> 38.230504).  Saving model ...
+    Epoch: 1 	Training Loss: 34.865927 	Validation Loss: 33.684661
+    Validation loss decreased (38.230504 --> 33.684661).  Saving model ...
+    Epoch: 2 	Training Loss: 30.692502 	Validation Loss: 29.130355
+    Validation loss decreased (33.684661 --> 29.130355).  Saving model ...
+    Epoch: 3 	Training Loss: 28.641049 	Validation Loss: 28.378742
+    Validation loss decreased (29.130355 --> 28.378742).  Saving model ...
+    Epoch: 4 	Training Loss: 27.138877 	Validation Loss: 26.230459
+    Validation loss decreased (28.378742 --> 26.230459).  Saving model ...
+    Epoch: 5 	Training Loss: 25.893545 	Validation Loss: 25.292490
+    Validation loss decreased (26.230459 --> 25.292490).  Saving model ...
+    Epoch: 6 	Training Loss: 24.842854 	Validation Loss: 24.485009
+    Validation loss decreased (25.292490 --> 24.485009).  Saving model ...
+    Epoch: 7 	Training Loss: 23.926138 	Validation Loss: 24.722069
+    Epoch: 8 	Training Loss: 23.031194 	Validation Loss: 23.820090
+    Validation loss decreased (24.485009 --> 23.820090).  Saving model ...
+    Epoch: 9 	Training Loss: 22.231426 	Validation Loss: 22.906112
+    Validation loss decreased (23.820090 --> 22.906112).  Saving model ...
+    Epoch: 10 	Training Loss: 21.461538 	Validation Loss: 22.387376
+    Validation loss decreased (22.906112 --> 22.387376).  Saving model ...
+    Epoch: 11 	Training Loss: 20.719964 	Validation Loss: 21.980570
+    Validation loss decreased (22.387376 --> 21.980570).  Saving model ...
+    Epoch: 12 	Training Loss: 20.001674 	Validation Loss: 21.879977
+    Validation loss decreased (21.980570 --> 21.879977).  Saving model ...
+    Epoch: 13 	Training Loss: 19.320499 	Validation Loss: 21.871980
+    Validation loss decreased (21.879977 --> 21.871980).  Saving model ...
+    Epoch: 14 	Training Loss: 18.759199 	Validation Loss: 21.631741
+    Validation loss decreased (21.871980 --> 21.631741).  Saving model ...
+    Epoch: 15 	Training Loss: 18.190729 	Validation Loss: 21.728372
+    Epoch: 16 	Training Loss: 17.555403 	Validation Loss: 21.431064
+    Validation loss decreased (21.631741 --> 21.431064).  Saving model ...
+    Epoch: 17 	Training Loss: 17.052087 	Validation Loss: 21.122314
+    Validation loss decreased (21.431064 --> 21.122314).  Saving model ...
+    Epoch: 18 	Training Loss: 16.523702 	Validation Loss: 21.020531
+    Validation loss decreased (21.122314 --> 21.020531).  Saving model ...
+    Epoch: 19 	Training Loss: 15.990338 	Validation Loss: 21.419434
+    Epoch: 20 	Training Loss: 15.540147 	Validation Loss: 22.715216
+    Epoch: 21 	Training Loss: 15.053955 	Validation Loss: 21.344116
+    Epoch: 22 	Training Loss: 14.620870 	Validation Loss: 21.724516
+    Epoch: 23 	Training Loss: 14.099778 	Validation Loss: 23.172254
+    Epoch: 24 	Training Loss: 13.740382 	Validation Loss: 22.565675
+    Epoch: 25 	Training Loss: 13.261968 	Validation Loss: 23.687115
+    Epoch: 26 	Training Loss: 12.842589 	Validation Loss: 23.510745
+    Epoch: 27 	Training Loss: 12.429801 	Validation Loss: 22.843567
+    Epoch: 28 	Training Loss: 12.002632 	Validation Loss: 24.907918
+    Epoch: 29 	Training Loss: 11.709404 	Validation Loss: 25.430181
+
 %% Cell type:markdown id:13e1df74 tags:
 
 Does overfit occur? If so, do an early stopping.
 
 %% Cell type:code id:d39df818 tags:
 
 ``` python
 import matplotlib.pyplot as plt
 
 plt.plot(range(n_epochs), train_loss_list)
 plt.xlabel("Epoch")
 plt.ylabel("Loss")
 plt.title("Performance of Model 1")
 plt.show()
 ```
 
+%% Output
+
+
+
 %% Cell type:markdown id:11df8fd4 tags:
 
 Now loading the model with the lowest validation loss value
 
 %% Cell type:code id:e93efdfc tags:
 
 ``` python
 model.load_state_dict(torch.load("./model_cifar.pt"))
 
 # track test loss
 test_loss = 0.0
 class_correct = list(0.0 for i in range(10))
 class_total = list(0.0 for i in range(10))
 
 model.eval()
 # iterate over test data
 for data, target in test_loader:
     # move tensors to GPU if CUDA is available
     if train_on_gpu:
         data, target = data.cuda(), target.cuda()
     # forward pass: compute predicted outputs by passing inputs to the model
     output = model(data)
     # calculate the batch loss
     loss = criterion(output, target)
     # update test loss
     test_loss += loss.item() * data.size(0)
     # convert output probabilities to predicted class
     _, pred = torch.max(output, 1)
     # compare predictions to true label
     correct_tensor = pred.eq(target.data.view_as(pred))
     correct = (
         np.squeeze(correct_tensor.numpy())
         if not train_on_gpu
         else np.squeeze(correct_tensor.cpu().numpy())
     )
     # calculate test accuracy for each object class
     for i in range(batch_size):
         label = target.data[i]
         class_correct[label] += correct[i].item()
         class_total[label] += 1
 
 # average test loss
 test_loss = test_loss / len(test_loader)
 print("Test Loss: {:.6f}\n".format(test_loss))
 
 for i in range(10):
     if class_total[i] > 0:
         print(
             "Test Accuracy of %5s: %2d%% (%2d/%2d)"
             % (
                 classes[i],
                 100 * class_correct[i] / class_total[i],
                 np.sum(class_correct[i]),
                 np.sum(class_total[i]),
             )
         )
     else:
         print("Test Accuracy of %5s: N/A (no training examples)" % (classes[i]))
 
 print(
     "\nTest Accuracy (Overall): %2d%% (%2d/%2d)"
     % (
         100.0 * np.sum(class_correct) / np.sum(class_total),
         np.sum(class_correct),
         np.sum(class_total),
     )
 )
 ```
 
+%% Output
+
+    Test Loss: 21.750090
+    
+    Test Accuracy of airplane: 68% (685/1000)
+    Test Accuracy of automobile: 75% (752/1000)
+    Test Accuracy of  bird: 51% (512/1000)
+    Test Accuracy of   cat: 37% (374/1000)
+    Test Accuracy of  deer: 61% (615/1000)
+    Test Accuracy of   dog: 57% (570/1000)
+    Test Accuracy of  frog: 68% (688/1000)
+    Test Accuracy of horse: 64% (646/1000)
+    Test Accuracy of  ship: 73% (736/1000)
+    Test Accuracy of truck: 73% (739/1000)
+    
+    Test Accuracy (Overall): 63% (6317/10000)
+
 %% Cell type:markdown id:944991a2 tags:
 
 Build a new network with the following structure.
 
 - It has 3 convolutional layers of kernel size 3 and padding of 1.
 - The first convolutional layer must output 16 channels, the second 32 and the third 64.
 - At each convolutional layer output, we apply a ReLU activation then a MaxPool with kernel size of 2.
 - Then, three fully connected layers, the first two being followed by a ReLU activation and a dropout whose value you will suggest.
 - The first fully connected layer will have an output size of 512.
 - The second fully connected layer will have an output size of 64.
 
 Compare the results obtained with this new network to those obtained previously.
 
+%% Cell type:code id: tags:
+
+``` python
+
+import torch.nn as nn
+import torch.nn.functional as F
+
+class CustomCNN(nn.Module):
+    def __init__(self, dropout_rate=0.3):  # Suggested dropout rate
+        super(CustomCNN, self).__init__()
+
+        # Convolutional layers
+        self.conv1 = nn.Conv2d(3, 16, 3, 1)
+        self.conv2 = nn.Conv2d(16, 32, 3, 1)
+        self.conv3 = nn.Conv2d(32, 64, 3, 1)
+
+        # Fully connected layers
+
+        self.fc1 = nn.Linear(64 * 2 * 2, 512)  # Corrigé en fonction de la forme réelle
+                      # Adjust input size based on input dimensions after pooling
+        self.fc2 = nn.Linear(512, 64)
+        self.fc3 = nn.Linear(64, 10)  # Assume 10 output classes
+
+        # Dropout
+        self.dropout = nn.Dropout(p=dropout_rate)
+
+        # Pooling
+        self.pool = nn.MaxPool2d(kernel_size=2)
+
+    def forward(self, x):
+        # Convolutional layers with ReLU and pooling
+        x = self.pool(F.relu(self.conv1(x)))
+        x = self.pool(F.relu(self.conv2(x)))
+        x = self.pool(F.relu(self.conv3(x)))
+
+        # Flatten for fully connected layers
+        x = x.view(x.size(0), -1)  # Flatten the tensor
+
+        # Fully connected layers with dropout and ReLU
+        x = self.dropout(F.relu(self.fc1(x)))
+        x = self.dropout(F.relu(self.fc2(x)))
+        x = self.fc3(x)
+
+        return x
+
+    # create a complete CNN
+model2 = CustomCNN()
+print(model2)
+# move tensors to GPU if CUDA is available
+if train_on_gpu:
+    model2.cuda()
+```
+
+%% Output
+
+    CustomCNN(
+      (conv1): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1))
+      (conv2): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1))
+      (conv3): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
+      (fc1): Linear(in_features=256, out_features=512, bias=True)
+      (fc2): Linear(in_features=512, out_features=64, bias=True)
+      (fc3): Linear(in_features=64, out_features=10, bias=True)
+      (dropout): Dropout(p=0.3, inplace=False)
+      (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
+    )
+
+%% Cell type:code id: tags:
+
+``` python
+print(f"Data type: {data.type()}")
+print(f"Target type: {target.type()}")
+print(f"Model weight type: {next(model2.parameters()).type()}")
+```
+
+%% Output
+
+    Data type: torch.cuda.FloatTensor
+    Target type: torch.cuda.LongTensor
+    Model weight type: torch.cuda.FloatTensor
+
+%% Cell type:code id: tags:
+
+``` python
+import torch.optim as optim
+
+criterion = nn.CrossEntropyLoss()  # specify loss function
+optimizer = optim.SGD(model2.parameters(), lr=0.01)  # specify optimizer
+
+n_epochs = 30  # number of epochs to train the model
+train_loss_list2 = []  # list to store loss to visualize
+valid_loss_min2 = np.Inf  # track change in validation loss
+
+for epoch in range(n_epochs):
+    # Keep track of training and validation loss
+    train_loss2 = 0.0
+    valid_loss2 = 0.0
+
+    # Train the model
+    model2.train()
+    for data, target in train_loader:
+        # Move tensors to GPU if CUDA is available
+        if train_on_gpu:
+            data, target = data.cuda(), target.cuda()
+        # Clear the gradients of all optimized variables
+        optimizer.zero_grad()
+        # Forward pass: compute predicted outputs by passing inputs to the model
+        output2 = model2(data)
+        # Calculate the batch loss
+        loss2 = criterion(output2, target)
+        # Backward pass: compute gradient of the loss with respect to model parameters
+        loss2.backward()
+        # Perform a single optimization step (parameter update)
+        optimizer.step()
+        # Update training loss
+        train_loss2 += loss2.item() * data.size(0)
+
+    # Validate the model
+    model2.eval()
+    with torch.no_grad():
+        for data, target in valid_loader:
+            if train_on_gpu:
+                data, target = data.cuda(), target.cuda()
+            else:
+                data, target = data.cpu(), target.cpu()
+            output2 = model2(data)
+            loss2 = criterion(output2, target)
+            valid_loss2 += loss2.item() * data.size(0)
+
+
+    # Calculate average losses
+    train_loss2 = train_loss2 / len(train_loader)
+    valid_loss2 = valid_loss2 / len(valid_loader)
+    train_loss_list2.append(train_loss2)
+
+    # Print training/validation statistics
+    print(
+        "Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}".format(
+            epoch, train_loss2, valid_loss2
+        )
+    )
+
+    # Save model if validation loss has decreased
+    if valid_loss2 <= valid_loss_min2:
+        print(
+            "Validation loss decreased ({:.6f} --> {:.6f}).  Saving model ...".format(
+                valid_loss_min2, valid_loss2
+            )
+        )
+        torch.save(model2.state_dict(), "model_cifar2.pt")
+        valid_loss_min2 = valid_loss2
+```
+
+%% Output
+
+    Epoch: 0 	Training Loss: 46.035675 	Validation Loss: 45.975821
+    Validation loss decreased (inf --> 45.975821).  Saving model ...
+    Epoch: 1 	Training Loss: 44.774520 	Validation Loss: 39.919449
+    Validation loss decreased (45.975821 --> 39.919449).  Saving model ...
+    Epoch: 2 	Training Loss: 37.829997 	Validation Loss: 34.754118
+    Validation loss decreased (39.919449 --> 34.754118).  Saving model ...
+    Epoch: 3 	Training Loss: 34.342419 	Validation Loss: 31.707682
+    Validation loss decreased (34.754118 --> 31.707682).  Saving model ...
+    Epoch: 4 	Training Loss: 31.886869 	Validation Loss: 29.301074
+    Validation loss decreased (31.707682 --> 29.301074).  Saving model ...
+    Epoch: 5 	Training Loss: 29.959563 	Validation Loss: 27.393316
+    Validation loss decreased (29.301074 --> 27.393316).  Saving model ...
+    Epoch: 6 	Training Loss: 28.335243 	Validation Loss: 26.011092
+    Validation loss decreased (27.393316 --> 26.011092).  Saving model ...
+    Epoch: 7 	Training Loss: 26.877296 	Validation Loss: 25.462102
+    Validation loss decreased (26.011092 --> 25.462102).  Saving model ...
+    Epoch: 8 	Training Loss: 25.688337 	Validation Loss: 23.929076
+    Validation loss decreased (25.462102 --> 23.929076).  Saving model ...
+    Epoch: 9 	Training Loss: 24.511723 	Validation Loss: 23.005517
+    Validation loss decreased (23.929076 --> 23.005517).  Saving model ...
+    Epoch: 10 	Training Loss: 23.442500 	Validation Loss: 21.846585
+    Validation loss decreased (23.005517 --> 21.846585).  Saving model ...
+    Epoch: 11 	Training Loss: 22.508246 	Validation Loss: 21.496349
+    Validation loss decreased (21.846585 --> 21.496349).  Saving model ...
+    Epoch: 12 	Training Loss: 21.597253 	Validation Loss: 20.689961
+    Validation loss decreased (21.496349 --> 20.689961).  Saving model ...
+    Epoch: 13 	Training Loss: 20.734205 	Validation Loss: 20.371868
+    Validation loss decreased (20.689961 --> 20.371868).  Saving model ...
+    Epoch: 14 	Training Loss: 19.955687 	Validation Loss: 19.954705
+    Validation loss decreased (20.371868 --> 19.954705).  Saving model ...
+    Epoch: 15 	Training Loss: 19.160272 	Validation Loss: 19.218016
+    Validation loss decreased (19.954705 --> 19.218016).  Saving model ...
+    Epoch: 16 	Training Loss: 18.510708 	Validation Loss: 18.665482
+    Validation loss decreased (19.218016 --> 18.665482).  Saving model ...
+    Epoch: 17 	Training Loss: 17.827416 	Validation Loss: 18.159658
+    Validation loss decreased (18.665482 --> 18.159658).  Saving model ...
+    Epoch: 18 	Training Loss: 17.263423 	Validation Loss: 18.637613
+    Epoch: 19 	Training Loss: 16.628417 	Validation Loss: 18.351383
+    Epoch: 20 	Training Loss: 16.099328 	Validation Loss: 18.048815
+    Validation loss decreased (18.159658 --> 18.048815).  Saving model ...
+    Epoch: 21 	Training Loss: 15.631801 	Validation Loss: 17.525234
+    Validation loss decreased (18.048815 --> 17.525234).  Saving model ...
+    Epoch: 22 	Training Loss: 15.120226 	Validation Loss: 17.490126
+    Validation loss decreased (17.525234 --> 17.490126).  Saving model ...
+    Epoch: 23 	Training Loss: 14.537110 	Validation Loss: 17.395222
+    Validation loss decreased (17.490126 --> 17.395222).  Saving model ...
+    Epoch: 24 	Training Loss: 14.215695 	Validation Loss: 18.225738
+    Epoch: 25 	Training Loss: 13.695718 	Validation Loss: 17.967939
+    Epoch: 26 	Training Loss: 13.258958 	Validation Loss: 18.192724
+    Epoch: 27 	Training Loss: 12.916577 	Validation Loss: 17.067720
+    Validation loss decreased (17.395222 --> 17.067720).  Saving model ...
+    Epoch: 28 	Training Loss: 12.522411 	Validation Loss: 16.906484
+    Validation loss decreased (17.067720 --> 16.906484).  Saving model ...
+    Epoch: 29 	Training Loss: 12.141524 	Validation Loss: 17.213408
+
+%% Cell type:code id: tags:
+
+``` python
+import matplotlib.pyplot as plt
+
+plt.plot(range(n_epochs), train_loss_list2)
+plt.xlabel("Epoch")
+plt.ylabel("Loss")
+plt.title("Performance of Model 2")
+plt.show()
+```
+
+%% Output
+
+
+
+%% Cell type:code id: tags:
+
+``` python
+model2.load_state_dict(torch.load("./model_cifar2.pt"))
+
+# track test loss
+test_loss2 = 0.0
+class_correct2 = list(0.0 for i in range(10))
+class_total2 = list(0.0 for i in range(10))
+
+model2.eval()
+# iterate over test data
+for data, target in test_loader:
+    # move tensors to GPU if CUDA is available
+    if train_on_gpu:
+        data, target = data.cuda(), target.cuda()
+    # forward pass: compute predicted outputs by passing inputs to the model
+    output = model2(data)
+    # calculate the batch loss
+    loss2 = criterion(output, target)
+    # update test loss
+    test_loss2 += loss2.item() * data.size(0)
+    # convert output probabilities to predicted class
+    _, pred = torch.max(output, 1)
+    # compare predictions to true label
+    correct_tensor = pred.eq(target.data.view_as(pred))
+    correct = (
+        np.squeeze(correct_tensor.numpy())
+        if not train_on_gpu
+        else np.squeeze(correct_tensor.cpu().numpy())
+    )
+    # calculate test accuracy for each object class
+    for i in range(batch_size):
+        label = target.data[i]
+        class_correct2[label] += correct[i].item()
+        class_total2[label] += 1
+
+# average test loss
+test_loss2 = test_loss2 / len(test_loader)
+print("Test Loss: {:.6f}\n".format(test_loss2))
+
+for i in range(10):
+    if class_total2[i] > 0:
+        print(
+            "Test Accuracy of %5s: %2d%% (%2d/%2d)"
+            % (
+                classes[i],
+                100 * class_correct2[i] / class_total2[i],
+                np.sum(class_correct2[i]),
+                np.sum(class_total2[i]),
+            )
+        )
+    else:
+        print("Test Accuracy of %5s: N/A (no training examples)" % (classes[i]))
+
+print(
+    "\nTest Accuracy (Overall): %2d%% (%2d/%2d)"
+    % (
+        100.0 * np.sum(class_correct2) / np.sum(class_total2),
+        np.sum(class_correct2),
+        np.sum(class_total2),
+    )
+)
+```
+
+%% Output
+
+    Test Loss: 17.564571
+    
+    Test Accuracy of airplane: 77% (776/1000)
+    Test Accuracy of automobile: 79% (793/1000)
+    Test Accuracy of  bird: 63% (639/1000)
+    Test Accuracy of   cat: 50% (506/1000)
+    Test Accuracy of  deer: 65% (655/1000)
+    Test Accuracy of   dog: 53% (536/1000)
+    Test Accuracy of  frog: 83% (835/1000)
+    Test Accuracy of horse: 70% (709/1000)
+    Test Accuracy of  ship: 80% (801/1000)
+    Test Accuracy of truck: 84% (841/1000)
+    
+    Test Accuracy (Overall): 70% (7091/10000)
+
 %% Cell type:markdown id:bc381cf4 tags:
 
 ## Exercise 2: Quantization: try to compress the CNN to save space
 
 Quantization doc is available from https://pytorch.org/docs/stable/quantization.html#torch.quantization.quantize_dynamic
 
 The Exercise is to quantize post training the above CNN model. Compare the size reduction and the impact on the classification accuracy
 
 
 The size of the model is simply the size of the file.
 
 %% Cell type:code id:ef623c26 tags:
 
 ``` python
 import os
 
 
 def print_size_of_model(model, label=""):
     torch.save(model.state_dict(), "temp.p")
     size = os.path.getsize("temp.p")
     print("model: ", label, " \t", "Size (KB):", size / 1e3)
     os.remove("temp.p")
     return size
 
 
 print_size_of_model(model, "fp32")
 ```
 
+%% Output
+
+    model:  fp32  	 Size (KB): 251.529
+
+    251529
+
 %% Cell type:markdown id:05c4e9ad tags:
 
 Post training quantization example
 
 %% Cell type:code id:c4c65d4b tags:
 
 ``` python
 import torch.quantization
 
 
 quantized_model = torch.quantization.quantize_dynamic(model, dtype=torch.qint8)
 print_size_of_model(quantized_model, "int8")
 ```
 
+%% Output
+
+    ---------------------------------------------------------------------------
+    RuntimeError                              Traceback (most recent call last)
+    <ipython-input-22-b8bd12179156> in <module>
+          2
+          3
+    ----> 4 quantized_model = torch.quantization.quantize_dynamic(model, dtype=torch.qint8)
+          5 print_size_of_model(quantized_model, "int8")
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/quantization/quantize.py in quantize_dynamic(model, qconfig_spec, dtype, mapping, inplace)
+        347     model.eval()
+        348     propagate_qconfig_(model, qconfig_spec)
+    --> 349     convert(model, mapping, inplace=True)
+        350     return model
+        351
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/quantization/quantize.py in convert(module, mapping, inplace, remove_qconfig)
+        412     if not inplace:
+        413         module = copy.deepcopy(module)
+    --> 414     _convert(module, mapping, inplace=True)
+        415     if remove_qconfig:
+        416         _remove_qconfig(module)
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/quantization/quantize.py in _convert(module, mapping, inplace)
+        457            not is_observed_custom_module(mod):
+        458             _convert(mod, mapping, inplace=True)
+    --> 459         reassign[name] = swap_module(mod, mapping)
+        460
+        461     for key, value in reassign.items():
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/quantization/quantize.py in swap_module(mod, mapping)
+        483             swapped = True
+        484         elif type(mod) in mapping:
+    --> 485             new_mod = mapping[type(mod)].from_float(mod)
+        486             swapped = True
+        487
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/nn/quantized/dynamic/modules/linear.py in from_float(cls, mod)
+         93         dtype = weight_observer.dtype
+         94         assert dtype in [torch.qint8, torch.float16], 'The only supported dtypes for dynamic quantized linear are qint8 and float16'
+    ---> 95         weight_observer(mod.weight)
+         96         if dtype == torch.qint8:
+         97             qweight = _quantize_weight(mod.weight.float(), weight_observer)
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
+        725             result = self._slow_forward(*input, **kwargs)
+        726         else:
+    --> 727             result = self.forward(*input, **kwargs)
+        728         for hook in itertools.chain(
+        729                 _global_forward_hooks.values(),
+    /opt/conda/envs/pytorch/lib/python3.8/site-packages/torch/quantization/observer.py in forward(self, x_orig)
+        391         x = x.to(self.min_val.dtype)
+        392         min_val_cur, max_val_cur = torch._aminmax(x)
+    --> 393         min_val = torch.min(min_val_cur, self.min_val)
+        394         max_val = torch.max(max_val_cur, self.max_val)
+        395         self.min_val.copy_(min_val)
+    RuntimeError: iter.device(arg).is_cuda() INTERNAL ASSERT FAILED at "/opt/conda/conda-bld/pytorch_1603729096996/work/aten/src/ATen/native/cuda/Loops.cuh":94, please report a bug to PyTorch.
+
 %% Cell type:markdown id:7b108e17 tags:
 
 For each class, compare the classification test accuracy of the initial model and the quantized model. Also give the overall test accuracy for both models.
 
+%% Cell type:code id: tags:
+
+``` python
+model2.cpu()
+quantized_model2 = torch.quantization.quantize_dynamic(model2, dtype=torch.qint8)
+print_size_of_model(quantized_model2, "int8")
+```
+
+%% Output
+
+    model:  int8  	 Size (KB): 267.281
+
+    267281
+
+%% Cell type:code id: tags:
+
+``` python
+# Après la quantification du modèle (avant d'enregistrer)
+quantized_model2 = quantized_model2.to('cpu')  # Déplace le modèle quantifié sur le CPU
+
+# Sauvegarder le modèle quantifié
+torch.save(quantized_model2.state_dict(), "quantized_model_cifar2.pt")
+
+# track test loss
+test_loss2q = 0.0
+class_correct2q = list(0.0 for i in range(10))
+class_total2q = list(0.0 for i in range(10))
+
+quantized_model2.eval()  # Modèle en mode évaluation
+# itérer sur les données de test
+for data, target in test_loader:
+    # Déplacer les données sur le CPU si elles sont sur le GPU
+    data, target = data.to('cpu'), target.to('cpu')  # Déplace les données et les cibles sur le CPU
+
+    # forward pass: calcul des sorties prédites en passant les entrées dans le modèle
+    output = quantized_model2(data)
+
+    # Calculer la perte par lot
+    loss2q = criterion(output, target)
+    test_loss2q += loss2q.item() * data.size(0)
+
+    # Convertir les probabilités de sortie en classes prédites
+    _, pred = torch.max(output, 1)
+
+    # Comparer les prédictions aux étiquettes vraies
+    correct_tensor = pred.eq(target.data.view_as(pred))
+    correct = (
+        np.squeeze(correct_tensor.numpy())  # Si non-CPU, utilisez CPU
+        if not train_on_gpu
+        else np.squeeze(correct_tensor.cpu().numpy())
+    )
+
+    # Calculer la précision de test pour chaque classe d'objet
+    for i in range(batch_size):
+        label = target.data[i]
+        class_correct2q[label] += correct[i].item()
+        class_total2q[label] += 1
+
+# Calcul de la perte moyenne de test
+test_loss2q = test_loss2q / len(test_loader)
+print("Test Loss: {:.6f}\n".format(test_loss2q))
+
+# Affichage de la précision pour chaque classe
+for i in range(10):
+    if class_total2q[i] > 0:
+        print(
+            "Test Accuracy of %5s: %2d%% (%2d/%2d)"
+            % (
+                classes[i],
+                100 * class_correct2q[i] / class_total2q[i],
+                np.sum(class_correct2q[i]),
+                np.sum(class_total2q[i]),
+            )
+        )
+    else:
+        print("Test Accuracy of %5s: N/A (no training examples)" % (classes[i]))
+
+# Affichage de la précision globale de test
+print(
+    "\nTest Accuracy (Overall): %2d%% (%2d/%2d)"
+    % (
+        100.0 * np.sum(class_correct2q) / np.sum(class_total2q),
+        np.sum(class_correct2q),
+        np.sum(class_total2q),
+    )
+)
+```
+
+%% Cell type:code id: tags:
+
+``` python
+
+torch.save(quantized_model2.state_dict(), "quantized_model_cifar2.pt")
+
+# track test loss
+test_loss2q = 0.0
+class_correct2q = list(0.0 for i in range(10))
+class_total2q = list(0.0 for i in range(10))
+
+quantized_model2.eval()
+# iterate over test data
+for data, target in test_loader:
+    # move tensors to GPU if CUDA is available
+    if train_on_gpu:
+        data, target = data.cuda(), target.cuda()
+    # forward pass: compute predicted outputs by passing inputs to the model
+    output = quantized_model2(data)
+    # calculate the batch loss
+    loss2q = criterion(output, target)
+    # update test loss
+    test_loss2q += loss2q.item() * data.size(0)
+    # convert output probabilities to predicted class
+    _, pred = torch.max(output, 1)
+    # compare predictions to true label
+    correct_tensor = pred.eq(target.data.view_as(pred))
+    correct = (
+        np.squeeze(correct_tensor.numpy())
+        if not train_on_gpu
+        else np.squeeze(correct_tensor.cpu().numpy())
+    )
+    # calculate test accuracy for each object class
+    for i in range(batch_size):
+        label = target.data[i]
+        class_correct2q[label] += correct[i].item()
+        class_total2q[label] += 1
+
+# average test loss
+test_loss2q = test_loss2q / len(test_loader)
+print("Test Loss: {:.6f}\n".format(test_loss2q))
+
+for i in range(10):
+    if class_total2q[i] > 0:
+        print(
+            "Test Accuracy of %5s: %2d%% (%2d/%2d)"
+            % (
+                classes[i],
+                100 * class_correct2q[i] / class_total2q[i],
+                np.sum(class_correct2q[i]),
+                np.sum(class_total2q[i]),
+            )
+        )
+    else:
+        print("Test Accuracy of %5s: N/A (no training examples)" % (classes[i]))
+
+print(
+    "\nTest Accuracy (Overall): %2d%% (%2d/%2d)"
+    % (
+        100.0 * np.sum(class_correct2q) / np.sum(class_total2q),
+        np.sum(class_correct2q),
+        np.sum(class_total2q),
+    )
+)
+```
+
+%% Cell type:raw id: tags:
+
+quantized_model.load_state_dict(torch.load("./model_cifarq.pt"))
+
+# track test loss
+test_lossq = 0.0
+class_correctq = list(0.0 for i in range(10))
+class_totalq = list(0.0 for i in range(10))
+
+quantized_model.eval()
+# iterate over test data
+for data, target in test_loader:
+    # move tensors to GPU if CUDA is available
+    if train_on_gpu:
+        data, target = data.cuda(), target.cuda()
+    # forward pass: compute predicted outputs by passing inputs to the model
+    output = quantized_model(data)
+    # calculate the batch loss
+    lossq = criterion(output, target)
+    # update test loss
+    test_lossq += lossq.item() * data.size(0)
+    # convert output probabilities to predicted class
+    _, pred = torch.max(output, 1)
+    # compare predictions to true label
+    correct_tensor = pred.eq(target.data.view_as(pred))
+    correctq = (
+        np.squeeze(correct_tensor.numpy())
+        if not train_on_gpu
+        else np.squeeze(correct_tensor.cpu().numpy())
+    )
+    # calculate test accuracy for each object class
+    for i in range(batch_size):
+        label = target.data[i]
+        class_correctq[label] += correctq[i].item()
+        class_totalq[label] += 1
+
+# average test loss
+test_lossq = test_lossq / len(test_loader)
+print("Test Loss: {:.6f}\n".format(test_lossq))
+
+for i in range(10):
+    if class_totalq[i] > 0:
+        print(
+            "Test Accuracy of %5s: %2d%% (%2d/%2d)"
+            % (
+                classes[i],
+                100 * class_correctq[i] / class_totalq[i],
+                np.sum(class_correctq[i]),
+                np.sum(class_totalq[i]),
+            )
+        )
+    else:
+        print("Test Accuracy of %5s: N/A (no training examples)" % (classes[i]))
+
+print(
+    "\nTest Accuracy (Overall): %2d%% (%2d/%2d)"
+    % (
+        100.0 * np.sum(class_correctq) / np.sum(class_totalq),
+        np.sum(class_correctq),
+        np.sum(class_totalq),
+    )
+)
+
 %% Cell type:markdown id:a0a34b90 tags:
 
 Try training aware quantization to mitigate the impact on the accuracy (doc available here https://pytorch.org/docs/stable/quantization.html#torch.quantization.quantize_dynamic)
 
 %% Cell type:markdown id:201470f9 tags:
 
 ## Exercise 3: working with pre-trained models.
 
 PyTorch offers several pre-trained models https://pytorch.org/vision/0.8/models.html
 We will use ResNet50 trained on ImageNet dataset (https://www.image-net.org/index.php). Use the following code with the files `imagenet-simple-labels.json` that contains the imagenet labels and the image dog.png that we will use as test.
 
 %% Cell type:code id:b4d13080 tags:
 
 ``` python
 import json
 from PIL import Image
 
 # Choose an image to pass through the model
 test_image = "dog.png"
 
 # Configure matplotlib for pretty inline plots
 #%matplotlib inline
 #%config InlineBackend.figure_format = 'retina'
 
 # Prepare the labels
 with open("imagenet-simple-labels.json") as f:
     labels = json.load(f)
 
 # First prepare the transformations: resize the image to what the model was trained on and convert it to a tensor
 data_transform = transforms.Compose(
     [
         transforms.Resize((224, 224)),
         transforms.ToTensor(),
         transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
     ]
 )
 # Load the image
 
 image = Image.open(test_image)
 plt.imshow(image), plt.xticks([]), plt.yticks([])
 
 # Now apply the transformation, expand the batch dimension, and send the image to the GPU
 # image = data_transform(image).unsqueeze(0).cuda()
 image = data_transform(image).unsqueeze(0)
 
 # Download the model if it's not there already. It will take a bit on the first run, after that it's fast
 model = models.resnet50(pretrained=True)
 # Send the model to the GPU
 # model.cuda()
 # Set layers such as dropout and batchnorm in evaluation mode
 model.eval()
 
 # Get the 1000-dimensional model output
 out = model(image)
 # Find the predicted class
 print("Predicted class is: {}".format(labels[out.argmax()]))
 ```
 
 %% Cell type:markdown id:184cfceb tags:
 
 Experiments:
 
 Study the code and the results obtained. Possibly add other images downloaded from the internet.
 
 What is the size of the model? Quantize it and then check if the model is still able to correctly classify the other images.
 
 Experiment with other pre-trained CNN models.
 
 
 
+%% Cell type:code id: tags:
+
+``` python
+import torch
+import torch.quantization
+from torchvision import models, transforms
+from PIL import Image
+import json
+import numpy as np
+import matplotlib.pyplot as plt
+
+# Prepare the transformations for the input image
+data_transform = transforms.Compose(
+    [
+        transforms.Resize((224, 224)),
+        transforms.ToTensor(),
+        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
+    ]
+)
+
+# Load the labels for ImageNet (a JSON file with label mappings)
+with open("imagenet-simple-labels.json") as f:
+    labels = json.load(f)
+
+# Choose an image to pass through the model
+test_image = "dog.png"
+
+# Load the image and apply transformations
+image = Image.open(test_image)
+plt.imshow(image)
+plt.xticks([]), plt.yticks([])  # Hide axis ticks
+plt.show()
+
+# Apply the data transformation and add a batch dimension
+image_tensor = data_transform(image).unsqueeze(0)
+
+# Load the pre-trained ResNet50 model
+model_resnet50 = models.resnet50(pretrained=True)
+model_resnet50.eval()  # Set model to evaluation mode
+
+# Quantize the model dynamically (8-bit integer)
+quantized_model_resnet50 = torch.quantization.quantize_dynamic(model_resnet50, dtype=torch.qint8)
+
+# Check the original and quantized model sizes
+def model_size(model):
+    return sum(p.numel() for p in model.parameters() if p.requires_grad) / 1e6  # In million parameters
+
+original_size = model_size(model_resnet50)
+quantized_size = model_size(quantized_model_resnet50)
+
+print(f"Original model size (ResNet50): {original_size:.2f} million parameters")
+print(f"Quantized model size (ResNet50): {quantized_size:.2f} million parameters")
+
+# Function to predict using the model
+def predict(model, image_tensor):
+    with torch.no_grad():  # No need to track gradients for inference
+        output = model(image_tensor)
+        _, predicted_class = torch.max(output, 1)
+        return labels[predicted_class.item()]
+
+# Test the original model (ResNet50)
+print("Original Model Prediction (ResNet50):", predict(model_resnet50, image_tensor))
+
+# Test the quantized model (ResNet50)
+quantized_model_resnet50.eval()  # Set to eval mode
+print("Quantized Model Prediction (ResNet50):", predict(quantized_model_resnet50, image_tensor))
+
+# Experiment with another model, e.g., AlexNet
+model_alexnet = models.alexnet(pretrained=True)
+model_alexnet.eval()
+print("AlexNet Prediction:", predict(model_alexnet, image_tensor))
+
+# Experiment with another model, e.g., VGG16
+model_vgg16 = models.vgg16(pretrained=True)
+model_vgg16.eval()
+print("VGG16 Prediction:", predict(model_vgg16, image_tensor))
+
+# You can add more models like ResNet18, DenseNet, etc., in a similar way.
+
+```
+
+%% Output
+
+
+
+    Original model size (ResNet50): 25.56 million parameters
+    Quantized model size (ResNet50): 23.51 million parameters
+    Original Model Prediction (ResNet50): Golden Retriever
+    Quantized Model Prediction (ResNet50): Golden Retriever
+
+    Downloading: "https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth" to /home/jcaty/.cache/torch/hub/checkpoints/alexnet-owt-4df8aa71.pth
+    70.5%IOPub message rate exceeded.
+    The notebook server will temporarily stop sending output
+    to the client in order to avoid crashing it.
+    To change this limit, set the config variable
+    `--NotebookApp.iopub_msg_rate_limit`.
+    
+    Current values:
+    NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
+    NotebookApp.rate_limit_window=3.0 (secs)
+    
+    8.3%IOPub message rate exceeded.
+    The notebook server will temporarily stop sending output
+    to the client in order to avoid crashing it.
+    To change this limit, set the config variable
+    `--NotebookApp.iopub_msg_rate_limit`.
+    
+    Current values:
+    NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
+    NotebookApp.rate_limit_window=3.0 (secs)
+    
+    33.0%IOPub message rate exceeded.
+    The notebook server will temporarily stop sending output
+    to the client in order to avoid crashing it.
+    To change this limit, set the config variable
+    `--NotebookApp.iopub_msg_rate_limit`.
+    
+    Current values:
+    NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)
+    NotebookApp.rate_limit_window=3.0 (secs)
+    
+
 %% Cell type:markdown id:5d57da4b tags:
 
 ## Exercise 4: Transfer Learning
 
 
 For this work, we will use a pre-trained model (ResNet18) as a descriptor extractor and will refine the classification by training only the last fully connected layer of the network. Thus, the output layer of the pre-trained network will be replaced by a layer adapted to the new classes to be recognized which will be in our case ants and bees.
 Download and unzip in your working directory the dataset available at the address :
 
 https://download.pytorch.org/tutorial/hymenoptera_data.zip
 
 Execute the following code in order to display some images of the dataset.
 
 %% Cell type:code id:be2d31f5 tags:
 
 ``` python
 import os
 
 import matplotlib.pyplot as plt
 import numpy as np
 import torch
 import torchvision
 from torchvision import datasets, transforms
 
 # Data augmentation and normalization for training
 # Just normalization for validation
 data_transforms = {
     "train": transforms.Compose(
         [
             transforms.RandomResizedCrop(
                 224
             ),  # ImageNet models were trained on 224x224 images
             transforms.RandomHorizontalFlip(),  # flip horizontally 50% of the time - increases train set variability
             transforms.ToTensor(),  # convert it to a PyTorch tensor
             transforms.Normalize(
                 [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]
             ),  # ImageNet models expect this norm
         ]
     ),
     "val": transforms.Compose(
         [
             transforms.Resize(256),
             transforms.CenterCrop(224),
             transforms.ToTensor(),
             transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
         ]
     ),
 }
 
 data_dir = "hymenoptera_data"
 # Create train and validation datasets and loaders
 image_datasets = {
     x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
     for x in ["train", "val"]
 }
 dataloaders = {
     x: torch.utils.data.DataLoader(
         image_datasets[x], batch_size=4, shuffle=True, num_workers=0
     )
     for x in ["train", "val"]
 }
 dataset_sizes = {x: len(image_datasets[x]) for x in ["train", "val"]}
 class_names = image_datasets["train"].classes
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 # Helper function for displaying images
 def imshow(inp, title=None):
     """Imshow for Tensor."""
     inp = inp.numpy().transpose((1, 2, 0))
     mean = np.array([0.485, 0.456, 0.406])
     std = np.array([0.229, 0.224, 0.225])
 
     # Un-normalize the images
     inp = std * inp + mean
     # Clip just in case
     inp = np.clip(inp, 0, 1)
     plt.imshow(inp)
     if title is not None:
         plt.title(title)
     plt.pause(0.001)  # pause a bit so that plots are updated
     plt.show()
 
 
 # Get a batch of training data
 inputs, classes = next(iter(dataloaders["train"]))
 
 # Make a grid from batch
 out = torchvision.utils.make_grid(inputs)
 
 imshow(out, title=[class_names[x] for x in classes])
 
 ```
 
 %% Cell type:markdown id:bbd48800 tags:
 
 Now, execute the following code which uses a pre-trained model ResNet18 having replaced the output layer for the ants/bees classification and performs the model training by only changing the weights of this output layer.
 
 %% Cell type:code id:572d824c tags:
 
 ``` python
 import copy
 import os
 import time
 
 import matplotlib.pyplot as plt
 import numpy as np
 import torch
 import torch.nn as nn
 import torch.optim as optim
 import torchvision
 from torch.optim import lr_scheduler
 from torchvision import datasets, transforms
 
 # Data augmentation and normalization for training
 # Just normalization for validation
 data_transforms = {
     "train": transforms.Compose(
         [
             transforms.RandomResizedCrop(
                 224
             ),  # ImageNet models were trained on 224x224 images
             transforms.RandomHorizontalFlip(),  # flip horizontally 50% of the time - increases train set variability
             transforms.ToTensor(),  # convert it to a PyTorch tensor
             transforms.Normalize(
                 [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]
             ),  # ImageNet models expect this norm
         ]
     ),
     "val": transforms.Compose(
         [
             transforms.Resize(256),
             transforms.CenterCrop(224),
             transforms.ToTensor(),
             transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
         ]
     ),
 }
 
 data_dir = "hymenoptera_data"
 # Create train and validation datasets and loaders
 image_datasets = {
     x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
     for x in ["train", "val"]
 }
 dataloaders = {
     x: torch.utils.data.DataLoader(
         image_datasets[x], batch_size=4, shuffle=True, num_workers=4
     )
     for x in ["train", "val"]
 }
 dataset_sizes = {x: len(image_datasets[x]) for x in ["train", "val"]}
 class_names = image_datasets["train"].classes
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 # Helper function for displaying images
 def imshow(inp, title=None):
     """Imshow for Tensor."""
     inp = inp.numpy().transpose((1, 2, 0))
     mean = np.array([0.485, 0.456, 0.406])
     std = np.array([0.229, 0.224, 0.225])
 
     # Un-normalize the images
     inp = std * inp + mean
     # Clip just in case
     inp = np.clip(inp, 0, 1)
     plt.imshow(inp)
     if title is not None:
         plt.title(title)
     plt.pause(0.001)  # pause a bit so that plots are updated
     plt.show()
 
 
 # Get a batch of training data
 # inputs, classes = next(iter(dataloaders['train']))
 
 # Make a grid from batch
 # out = torchvision.utils.make_grid(inputs)
 
 # imshow(out, title=[class_names[x] for x in classes])
 # training
 
 
 def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
     since = time.time()
 
     best_model_wts = copy.deepcopy(model.state_dict())
     best_acc = 0.0
 
     epoch_time = []  # we'll keep track of the time needed for each epoch
 
     for epoch in range(num_epochs):
         epoch_start = time.time()
         print("Epoch {}/{}".format(epoch + 1, num_epochs))
         print("-" * 10)
 
         # Each epoch has a training and validation phase
         for phase in ["train", "val"]:
             if phase == "train":
                 scheduler.step()
                 model.train()  # Set model to training mode
             else:
                 model.eval()  # Set model to evaluate mode
 
             running_loss = 0.0
             running_corrects = 0
 
             # Iterate over data.
             for inputs, labels in dataloaders[phase]:
                 inputs = inputs.to(device)
                 labels = labels.to(device)
 
                 # zero the parameter gradients
                 optimizer.zero_grad()
 
                 # Forward
                 # Track history if only in training phase
                 with torch.set_grad_enabled(phase == "train"):
                     outputs = model(inputs)
                     _, preds = torch.max(outputs, 1)
                     loss = criterion(outputs, labels)
 
                     # backward + optimize only if in training phase
                     if phase == "train":
                         loss.backward()
                         optimizer.step()
 
                 # Statistics
                 running_loss += loss.item() * inputs.size(0)
                 running_corrects += torch.sum(preds == labels.data)
 
             epoch_loss = running_loss / dataset_sizes[phase]
             epoch_acc = running_corrects.double() / dataset_sizes[phase]
 
             print("{} Loss: {:.4f} Acc: {:.4f}".format(phase, epoch_loss, epoch_acc))
 
             # Deep copy the model
             if phase == "val" and epoch_acc > best_acc:
                 best_acc = epoch_acc
                 best_model_wts = copy.deepcopy(model.state_dict())
 
         # Add the epoch time
         t_epoch = time.time() - epoch_start
         epoch_time.append(t_epoch)
         print()
 
     time_elapsed = time.time() - since
     print(
         "Training complete in {:.0f}m {:.0f}s".format(
             time_elapsed // 60, time_elapsed % 60
         )
     )
     print("Best val Acc: {:4f}".format(best_acc))
 
     # Load best model weights
     model.load_state_dict(best_model_wts)
     return model, epoch_time
 
 
 # Download a pre-trained ResNet18 model and freeze its weights
 model = torchvision.models.resnet18(pretrained=True)
 for param in model.parameters():
     param.requires_grad = False
 
 # Replace the final fully connected layer
 # Parameters of newly constructed modules have requires_grad=True by default
 num_ftrs = model.fc.in_features
 model.fc = nn.Linear(num_ftrs, 2)
 # Send the model to the GPU
 model = model.to(device)
 # Set the loss function
 criterion = nn.CrossEntropyLoss()
 
 # Observe that only the parameters of the final layer are being optimized
 optimizer_conv = optim.SGD(model.fc.parameters(), lr=0.001, momentum=0.9)
 exp_lr_scheduler = lr_scheduler.StepLR(optimizer_conv, step_size=7, gamma=0.1)
 model, epoch_time = train_model(
     model, criterion, optimizer_conv, exp_lr_scheduler, num_epochs=10
 )
 ```
 
-%% Cell type:markdown id:bbd48800 tags:
+%% Cell type:markdown id:ca7cda74 tags:
 
 Experiments:
 Study the code and the results obtained.
 
 Modify the code and add an "eval_model" function to allow
 the evaluation of the model on a test set (different from the learning and validation sets used during the learning phase). Study the results obtained.
 
 Now modify the code to replace the current classification layer with a set of two layers using a "relu" activation function for the middle layer, and the "dropout" mechanism for both layers. Renew the experiments and study the results obtained.
 
 Apply ther quantization (post and quantization aware) and evaluate impact on model size and accuracy.
 
 %% Cell type:markdown id:04a263f0 tags:
 
 ## Optional
 
 Try this at home!!
 
 
 Pytorch offers a framework to export a given CNN to your selfphone (either android or iOS). Have a look at the tutorial https://pytorch.org/mobile/home/
 
 The Exercise consists in deploying the CNN of Exercise 4 in your phone and then test it on live.
 
 
 %% Cell type:markdown id:fe954ce4 tags:
 
 ## Author
 
 Alberto BOSIO - Ph. D.
+
+%% Cell type:markdown id: tags:
+
+## Author
+
+Alberto BOSIO - Ph. D.
+
+%% Cell type:markdown id: tags:
+
+## Author
+
+Alberto BOSIO - Ph. D.