Tutorial: PyTorch Example Code

Preparation

1. Create the working directory: such as \TU_MLP_CNN_Classification\

2. Download the Tutorial Code in Zip file: TU_MLP_CNN_Classification_Example Code 2024.zip

3. Unzip the files in the working directory.

4. Make sure you have the subdirectory for saving weights: \TU_MLP_CNN_Classification\weights\

Example 1. ANN(MLP) : Model, Train, Test

Classify the MNIST digit with a simple ANN\

• Image Input: 1x28x28 image

• Flatten into a 1x28*28 element vector

• 1st Layer: linear to 250 dimensions / ReLU

• 2nd Layer: linear to 100 dimensions / ReLU

• 3rd Layer: linear to 10 dimensions / log SoftMax

• Output: 1x10

• Activation function: ReLU

Creating Model Architecture, Training and Evaluation

We will create a class for the model architecture

Also, we will create PyTorch source files for processing (1) Training and (2) Evaluation of the model

1. Check if you have created the subfolder .\weights\

2. Download the source code for Model Architecture: myModel.py

3. Download the source code for Model Training: TU_PyTorch_MLP_Train.py

   • Run this code and observe the loss

   • Change the training epoch

4. Download the source code for Model Evaluation: TU_PyTorch_MLP_Eval.py

   • Run this code and observe the evaluation accuracy and sample images

myModel.py

```python
##########################################################
# PyTorch Tutorial:  MLP & CNN Model Architecture
#
# Author: Y.K.Kim
# mod: 2024-5-21 
#
# Descr: User defined ANN & CNN Model & Modules for Training and Testing
#
# Model: MLP, LeNet
#
##########################################################

import torch 
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor
import numpy as np 
import matplotlib.pyplot as plt




##########################################################
##  Model Architecture   
##########################################################


# Model Architecture: MLP
class MLP(nn.Module):
    def __init__(self):
        super(MLP, self).__init__()
        self.flatten = nn.Flatten()
        self.linear1 = nn.Linear(28*28, 250)
        self.linear2 = nn.Linear(250, 100)
        self.linear3 = nn.Linear(100, 10)

        
    def forward(self, x):
        x=self.flatten(x)
        x= F.relu(self.linear1(x))
        x = F.relu(self.linear2(x))
        y_pred = F.log_softmax(self.linear3(x))
        return y_pred


     

##########################################################
##  Train Module
##########################################################

# Train Module
def train(dataloader, model, loss_fn,optimizer, device, print_interval=100):
    # Dataset Size
    size = len(dataloader.dataset)
    
    # Batch size
    batch_size = dataloader.batch_size
    print(f"batch size : {batch_size}")

    # Model in Training Mode
    model.train()

    running_loss=0.0

    for batch, (X, y) in enumerate(dataloader):
        X, y = X.to(device), y.to(device)
        
        # zero gradients for every batch
        optimizer.zero_grad()   

        # Compute prediction loss 
        pred = model(X)
        loss = loss_fn(pred, y)

        # Backpropagation and Update        
        loss.backward()
        optimizer.step()        

        # Print avg. loss for every mini-batch in an epoch
        running_loss+=loss.item()
        if batch % print_interval == 0:
            running_loss=running_loss/print_interval            
            current = batch * batch_size
            print(f"loss: {running_loss:>7f}  [{current:>5d}/{size:>5d}]")
            running_loss=0




##########################################################
##  Test Model Module 
##########################################################

def test(dataloader, model, device):
    # Dataset Size
    size = len(dataloader.dataset)

    # Batch Size
    num_batches = len(dataloader)
    
    # Model in Evaluation Mode
    model.eval()

    #test_loss=0
    correctN = 0
    
    # Disable grad() computation to reduce memory consumption.
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            
            # Compute average prediction loss 
            pred = model(X)            
            #test_loss += loss_fn(pred, y).item()

            # Predict Label
            y_pred=pred.argmax(1);
            correctN += (y_pred == y).type(torch.float).sum().item()
            
    #test_loss /= num_batches
    correctN /= size
    #print(f"Test Error: \n Accuracy: {(100*correctN):>0.1f}%, Avg loss: {test_loss:>8f} \n")
    print(f"Test Error: \n Accuracy: {(100*correctN):>0.1f}% \n")


```

TU_PyTorch_MLP_Train

TU_PyTorch_MLP_Eval

(For CoLAB) Everything in one source file

(For CoLAB) Everything in one source file

```python
##########################################################
# PyTorch Tutorial:  Classification MLP Model for Evaluation
#
# Author: Y.K.Kim
# mod: 2024-5-21 
#
# Descr: This example is tesing pretrined model for classification
#
# Model:   MLP
# Dataset: MNIST
#
##########################################################
```

import torch 
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor
import numpy as np 
import matplotlib.pyplot as plt

# Select GPU or CPU for training.
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using {device} device")


##########################################################
## Part 1:  Prepare Dataset
##########################################################

# Download Dataset from TorchVision MNIST
# Once, downloaded locally, it does not download again.
training_data = datasets.MNIST(
    root="data",
    train=True,
    download=True,
    transform=ToTensor(),   #converts 0~255 value to 0~1 value.
)
test_data = datasets.MNIST(
    root="data",
    train=False,
    download=True,
    transform=ToTensor(),
)

# Create DataLoader with Batch size N
batch_size = 64
train_dataloader = DataLoader(training_data, batch_size=batch_size, shuffle=True)
test_dataloader = DataLoader(test_data, batch_size=batch_size, shuffle=True)

for X, y in test_dataloader:
    print(f"Shape of X [N, C, H, W]: {X.shape}")
    print(f"Shape of y: {y.shape} {y.dtype}")
    break


##########################################################
## Part 2:  Create Model
##########################################################

# Model Architecture
class MLP(nn.Module):
    def __init__(self):
        super(MLP, self).__init__()
        self.flatten = nn.Flatten()
        self.linear1 = nn.Linear(28*28, 250)
        self.linear2 = nn.Linear(250, 100)
        self.linear3 = nn.Linear(100, 10)

    def forward(self, x):
        x=self.flatten(x)
        x= F.relu(self.linear1(x))
        x = F.relu(self.linear2(x))
        y_pred = F.log_softmax(self.linear3(x))
        return y_pred

# Model Class
model = MLP().to(device)
print(model)

# Loss Function
loss_fn = nn.CrossEntropyLoss()

# Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=1e-3)


##########################################################
## Part 3:  Train Model
##########################################################

# Train Module
def train(dataloader, model, loss_fn, optimizer):
    # Dataset Size
    size = len(dataloader.dataset)
    
    # Model in Training Mode
    model.train()

    running_loss=0.0

    for batch, (X, y) in enumerate(dataloader):
        X, y = X.to(device), y.to(device)

        # zero gradients for every batch
        optimizer.zero_grad()   

        # Compute prediction loss 
        pred = model(X)
        loss = loss_fn(pred, y)

        # Backpropagation and Update        
        loss.backward()
        optimizer.step()        

        # Print loss for every 100 batch in an epoch
        running_loss+=loss.item()
        if batch % 100 == 0:
            running_loss=running_loss/100
            current = batch * len(X)
            print(f"loss: {running_loss:>7f}  [{current:>5d}/{size:>5d}]")
            running_loss=0

# Run Train for k epoch
epochs = 2
for t in range(epochs):
    print(f"Epoch {t+1}\n-------------------------------")
    train(train_dataloader, model, loss_fn, optimizer)    
print("Done!")

# Save Train Model
torch.save(model,"MNIST_model.pth")


##########################################################
## Part 4:  Test Model - Evaluation
##########################################################

def test(dataloader, model, loss_fn):
    # Dataset Size
    size = len(dataloader.dataset)

    # Batch Size
    num_batches = len(dataloader)
    
    # Model in Evaluation Mode
    model.eval()

    test_loss, correctN = 0, 0
    
    # Disable grad() computation to reduce memory consumption.
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            
            # Compute average prediction loss 
            pred = model(X)            
            test_loss += loss_fn(pred, y).item()

            # Predict Label
            y_pred=pred.argmax(1);
            correctN += (y_pred == y).type(torch.float).sum().item()
            
    test_loss /= num_batches
    correctN /= size
    print(f"Test Error: \n Accuracy: {(100*correctN):>0.1f}%, Avg loss: {test_loss:>8f} \n")

# Run Eval for k epoch
test(test_dataloader, model, loss_fn)


##########################################################
## Part 5:  Visualize Evaluation Results
##########################################################

# Select one batch of images
dataiter = iter(test_dataloader)
images, labels = next(dataiter)
print(images.size())

# Evaluate on one batch test images
with torch.no_grad():
  for X, y in dataiter:
      X, y = X.to(device), y.to(device)
      
      # Prediction Label 
      pred = model(X)
      _, predicted = torch.max(pred.data, 1)

# Plot 
figure = plt.figure()
num_of_images = 9
for index in range(num_of_images):
    plt.subplot(3, 3, index+1)
    plt.axis('off')    
    plt.title("Predicted: {}".format(predicted[index].item()))
    plt.imshow(images[index].cpu().numpy().squeeze(), cmap='gray_r')
    plt.show()

Example 2. CNN : Model, Train, Test

LeNet-5 Architecture. Credit: LeCun et al., 1998

Architecture

• [C1] Conv: Input (32x32x1) to Output (28x28x6) by 5x5 filters, relu

• [S2] Pooling : Input (28x28x6) to Output (14x14x6), maxPooling 2x2

• [C3] Conv: Input (14x14x6) to Output (10x10x16) by 5x5 filters, relu

• [S4] Pooling : Input (10x10x16) to Output (5x5x16), maxPooling 2x2

• Flatten : Input (5x5x16) to Output 1D 1x (5 * 5 * 16)

• [F5] FC : Input (1x5 * 5 * 16) to Output (1x120) , relu

• [F6] FC : Input (1x120) to Output (1x84) , relu

• [OUTPUT] : Input (1x84) to Output (1x10)

Creating Model Architecture, Training and Evaluation

We will worki in the same working directory as Example 1: e.g. \TU_MLP_CNN_Classification\

We will create a class for the model architecture. Also, we will create PyTorch source files for processing (1) Training and (2) Evaluation of the model

1. Update myModel.py by adding class LeNet5

2. Download the source code for Model Training: TU_PyTorch_CNN_Train.py

   • Run this code and observe the loss

   • Change the training epoch

3. Download the source code for Model Evaluation: TU_PyTorch_CNN_Eval.py

   • Run this code and observe the evaluation accuracy and sample images

Other Style using nn.Sequencial()

Important Note

• in Sequential(). use nn.ReLU() , NOT F.relu()

• ( nn.ReLU() vs F.relu() ) Watch out for spelling and Captital Letters

• add comma , in nn.Sequential(). Such as nn.Conv2d(1,6,5), nn.ReLU() etc

myModel.py

# myModel.py

# ... same as Example 1

# Model Architecture
class LeNet5(nn.Module):

    def __init__(self):
        super(LeNet5, self).__init__()
        self.flatten = nn.Flatten()

        # Feature Extraction 
        # Conv2d(input ch, output ch, convolution,stride=1)
        # (1,6,5) = Input ch=1, Output ch=6 with 5x5 conv kernel        
        self.conv1 = nn.Conv2d(1, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
                
        # Classifier
        self.fc1 = nn.Linear(16*5*5,120) 
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):        
        # C1
        x=self.conv1(x)
        x=F.relu(x)
        # S2 MaxPool 2x2
        x = F.max_pool2d(x, 2)                
        # C3
        x = F.relu(self.conv2(x))
        # S4 
        x = F.max_pool2d(x, 2)
        # Flatten              
        x = self.flatten(x)
        # F5
        x = F.relu(self.fc1(x))        
        # F6 
        x = F.relu(self.fc2(x))
        # OUTPUT
        logits = self.fc3(x)

        probs = F.softmax(logits,dim=1) # y_pred: 0~1 
        return probs

myModel.py (2nd version)

The only changes in Example 2 are (1) creating CNN model class (2) changing the input image dimension. The rest of the code are the same

Train and Evaluation of CNN Classification

LeNet uses 1x32x32 input. Therefore, reshape MNIST from 1x28x28 to 1x32x32

• MLP uses 1x28x28 as the Input for MNIST

• LeNet uses 1x32x32 as the Input for MNIST

1. Download the source code for Model Training: TU_PyTorch_CNN_Train.py

   • Run this code and observe the loss

   • Change the training epoch

2. Download the source code for Model Evaluation: TU_PyTorch_CNN_Eval.py

   • Run this code and observe the evaluation accuracy and sample images

TU_PyTorch_CNN_Train.py

```python
##########################################################
# PyTorch Tutorial:  Classification CNN Model for Training
#
# Author: Y.K.Kim
# mod: 2024-5-21 
#
# Descr: This example is creating and training a CNN model for classification
#
# Model: LeNet
# Dataset: MNIST
#
##########################################################


import torch 
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor
import numpy as np 
import matplotlib.pyplot as plt

# User defined Model
import myModel  as myModel 

# Model weight directory
ModelPATH='./weights/'

##########################################################
## Part 0:  GPU setting
##########################################################

# Select GPU or CPU for training.
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using {device} device")



##########################################################
## Part 1:  Prepare Dataset 
##########################################################

# Download Dataset from TorchVision MNIST
# Once, downloaded locally, it does not download again.
#
# NOTE: LeNet uses 1x32x32 input. 
# Reshape MNIST from 1x28x28 to  1x32x32

# transformation for resize
data_transform = transforms.Compose([
            transforms.Resize((32, 32)),
            transforms.ToTensor(),
])

# TRAIN DATA
training_data = datasets.MNIST(
    root="data",
    train=True,
    download=True,
    transform=data_transform,   #converts data size
)


# Create DataLoader with Batch size N
# Input Dim:  [N, C, H, W]
batch_size = 64
train_dataloader = DataLoader(training_data, batch_size=batch_size, shuffle=True)


for X, y in train_dataloader:
    print(f"Shape of X [N, C, H, W]: {X.shape}")
    print(f"Shape of y: {y.shape} {y.dtype}")
    break




##########################################################
## Part 2:  Create Model Instance - CNN
##########################################################

# Model Class Construction
model = myModel.LeNet5().to(device)
print(model)



##########################################################
## Part 3:  Train Model
##########################################################

# Loss Function
loss_fn = nn.CrossEntropyLoss()

# Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=1e-3)

# Run Train for k epoch
epochs = 1
for t in range(epochs):
    print(f"Epoch {t+1}\n-------------------------------")
    myModel.train(train_dataloader, model,loss_fn, optimizer,device)    
print("Done!")


# Save Train Model
# * Need to create a new folder PATH priorly
torch.save(model,ModelPATH+'LeNet_MNIST_model.pth')



# Evaluate the trained model in **Eval.py
```

TU_PyTorch_CNN_Eval.py