#1 Basics

🔥 PyTorch Hands-On Course for Beginners

10 practical modules with code, exercises, and assignments.

Module 1: PyTorch Basics

Explanation:
This module introduces PyTorch tensors, the fundamental data structure used to store data in PyTorch. You learn how to create tensors, inspect their shape, and understand data types.

Learning Outcomes:

• Understand what a tensor is
• Create tensors using PyTorch
• Inspect tensor shape and data type

Create your first tensor.

                    import torch

                    x = torch.tensor([1, 2, 3])
                    print(x)
                    print(x.dtype)
                    print(x.shape)
                    

Exercise,Assignment & MCQ

Exercises

• Create a tensor with values [10, 20, 30]
• Print tensor shape and data type
• Create a 2×2 zero tensor

Assignment

Create a tensor representing marks of 5 students and compute the average.

MCQ Quiz

• 1. What is the core data structure in PyTorch?
  A. List
  B. Array
  C. Tensor
  D. Matrix
• 2. Which function creates a tensor?
  A. torch.create()
  B. torch.tensor()
  C. torch.make()
  D. torch.build()
• 3. What does x.shape return?
  A. Data type
  B. Dimensions of tensor
  C. Memory size
  D. Values
• 4. PyTorch is mainly used for?
  A. Web design
  B. Deep learning
  C. OS development
  D. Gaming
• 5. Which library supports GPU acceleration?
  A. Pandas
  B. NumPy
  C. PyTorch
  D. Matplotlib

Module 2: Tensor Operations

Explanation:
This module covers basic mathematical operations on tensors such as addition, multiplication, and dot products using PyTorch.

Learning Outcomes:

• Perform arithmetic operations on tensors
• Apply element-wise operations
• Understand how tensors behave mathematically

                
                    import torch

                    a = torch.tensor([2, 4])
                    b = torch.tensor([1, 3])

                    print(a + b)
                    print(a * b)
                    print(torch.dot(a, b))
                
            

Exercises

• Subtract tensors
• Compute mean

Assignment

Simulate monthly expenses using tensors.

MCQ Quiz

• 1. Which operator performs element-wise addition?
  A. +
  B. dot()
  C. sum()
  D. mean()
• 2. What does torch.dot() compute?
  A. Element-wise product
  B. Dot product
  C. Matrix multiplication
  D. Sum
• 3. Which operation multiplies tensors element-wise?
  A. torch.dot()
  B. *
  C. torch.mul_all()
  D. add()
• 4. Tensor operations are?
  A. Scalar only
  B. Slow
  C. Vectorized
  D. Manual
• 5. Which function calculates average?
  A. torch.sum()
  B. torch.mean()
  C. torch.avg()
  D. torch.calc()

Module 3: Autograd

Explanation:
This module explains PyTorch’s autograd system, which automatically computes gradients required for training neural networks.

Learning Outcomes:

• Understand automatic differentiation
• Compute gradients using backward()
• Explain the role of gradients in learning

import torch

x = torch.tensor(2.0, requires_grad=True)
y = x**3 + 4*x
y.backward()
print(x.grad)

Exercises

• Compute gradient of y = x² + 5x
• Change value of x and observe gradient

Assignment

Derive gradient for a cubic equation using autograd.

MCQ Quiz

• 1. What does autograd compute?
  A. Values
  B. Gradients
  C. Shapes
  D. Loss
• 2. Which flag enables gradient tracking?
  A. grad=True
  B. requires_grad=True
  C. track=True
  D. backward=True
• 3. Which method computes gradients?
  A. grad()
  B. backward()
  C. step()
  D. compute()
• 4. Gradients are used in?
  A. Evaluation
  B. Optimization
  C. Data loading
  D. Saving
• 5. Gradients are stored in?
  A. x.value
  B. x.grad
  C. x.data
  D. x.diff

Module 4: Linear Regression

Explanation:
In this module, you build a simple linear regression model using PyTorch’s built-in layers, loss functions, and optimizers.

Learning Outcomes:

• Build a linear regression model
• Use loss functions and optimizers
• Train a model using gradient descent

import torch

x = torch.tensor([[1.],[2.],[3.]])
y = torch.tensor([[2.],[4.],[6.]])

model = torch.nn.Linear(1,1)
loss_fn = torch.nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

for _ in range(200):
    loss = loss_fn(model(x), y)
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

print(model.weight.item(), model.bias.item())

Exercises

• Change learning rate and observe loss
• Increase training epochs

Assignment

Train a model to predict salary based on years of experience.

MCQ Quiz

• 1. Linear regression predicts?
  A. Categories
  B. Continuous values
  C. Images
  D. Text
• 2. Which loss is used for regression?
  A. BCELoss
  B. MSELoss
  C. CrossEntropy
  D. L1Loss
• 3. Which optimizer updates weights?
  A. Loss
  B. Optimizer
  C. Dataset
  D. Model
• 4. Learning rate controls?
  A. Data size
  B. Step size
  C. Accuracy
  D. Output
• 5. Which layer is used?
  A. nn.Linear
  B. nn.Conv2d
  C. nn.RNN
  D. nn.Dropout

Module 5: Dataset & DataLoader

Explanation:
This module introduces Dataset and DataLoader classes to efficiently load data in batches during training.

Learning Outcomes:

• Create a custom Dataset class
• Use DataLoader for batching
• Understand data iteration during training

from torch.utils.data import Dataset, DataLoader
import torch

class MyData(Dataset):
    def __init__(self):
        self.x = torch.arange(1,11).float()
        self.y = self.x * 2

    def __len__(self):
        return len(self.x)

    def __getitem__(self, i):
        return self.x[i], self.y[i]

loader = DataLoader(MyData(), batch_size=2)
for x,y in loader:
    print(x,y)

Exercises

• Change batch size to 5
• Add random noise to dataset

Assignment

Create a Dataset class for student marks and grades.

MCQ Quiz

• 1. Dataset class is used to?
  A. Store data
  B. Train model
  C. Optimize loss
  D. Save model
• 2. DataLoader provides?
  A. Prediction
  B. Batching
  C. Loss
  D. Accuracy
• 3. __getitem__ returns?
  A. Length
  B. One sample
  C. All data
  D. Batch size
• 4. DataLoader improves?
  A. Accuracy
  B. Efficiency
  C. Loss
  D. Weights
• 5. Batch size controls?
  A. Epochs
  B. Samples per iteration
  C. Learning rate
  D. Layers

Module 6: Neural Network

Explanation:
This module demonstrates how to construct a multi-layer neural network using linear layers and activation functions.

Learning Outcomes:

• Understand neural network architecture
• Create hidden layers
• Apply activation functions

import torch.nn as nn

model = nn.Sequential(
    nn.Linear(2,4),
    nn.ReLU(),
    nn.Linear(4,1)
)
print(model)

Exercises

• Add another hidden layer
• Change activation function to Tanh

Assignment

Design a neural network with 3 hidden layers.

MCQ Quiz

• 1. Neural networks consist of?
  A. Layers
  B. Tables
  C. Loops
  D. Files
• 2. ReLU is a?
  A. Loss
  B. Activation function
  C. Optimizer
  D. Dataset
• 3. Hidden layers increase?
  A. Model capacity
  B. Data size
  C. Speed
  D. Storage
• 4. nn.Sequential is used to?
  A. Stack layers
  B. Train model
  C. Load data
  D. Save weights
• 5. Output layer size depends on?
  A. Problem type
  B. Dataset size
  C. Epochs
  D. Optimizer

Module 7: Training Loop

Explanation:
This module explains the standard training loop used in PyTorch, including forward pass, loss computation, backward pass, and parameter updates.

Learning Outcomes:

• Implement a complete training loop
• Understand model parameter updates
• Monitor training progress

import torch, torch.nn as nn

x = torch.randn(100,2)
y = x.sum(dim=1, keepdim=True)

model = nn.Linear(2,1)
opt = torch.optim.Adam(model.parameters(), lr=0.01)
loss_fn = nn.MSELoss()

for _ in range(100):
    loss = loss_fn(model(x), y)
    opt.zero_grad()
    loss.backward()
    opt.step()

Exercises

• Print loss every 10 epochs
• Change optimizer to SGD

Assignment

Implement early stopping logic.

MCQ Quiz

• 1. First step in training loop?
  A. Forward pass
  B. Backward pass
  C. Save model
  D. Eval
• 2. backward() computes?
  A. Gradients
  B. Loss
  C. Output
  D. Accuracy
• 3. optimizer.step() does?
  A. Update weights
  B. Reset loss
  C. Load data
  D. Print output
• 4. optimizer.zero_grad() is used to?
  A. Clear gradients
  B. Save memory
  C. Increase speed
  D. Stop training
• 5. Epoch means?
  A. One full dataset pass
  B. One batch
  C. One layer
  D. One sample

Module 8: Evaluation

Explanation:
This module focuses on evaluating trained models by switching to evaluation mode and disabling gradient computation.

Learning Outcomes:

• Use model.eval() correctly
• Disable gradients during evaluation
• Run inference safely

model.eval()
with torch.no_grad():
    print(model(torch.tensor([[3.0,5.0]])))

Exercises

• Test model with multiple inputs
• Compare outputs in train vs eval mode

Assignment

Write evaluation code for accuracy calculation.

MCQ Quiz

• 1. model.eval() is used for?
  A. Evaluation
  B. Training
  C. Saving
  D. Loading
• 2. torch.no_grad() does?
  A. Disables gradients
  B. Speeds GPU
  C. Clears memory
  D. Stops training
• 3. Evaluation mode affects?
  A. Dropout & BatchNorm
  B. Loss only
  C. Optimizer
  D. Dataset
• 4. Evaluation should be done?
  A. After training
  B. Before training
  C. During loading
  D. During saving
• 5. Gradients are needed during eval?
  A. No
  B. Yes
  C. Sometimes
  D. Always

Module 9: Save & Load

Explanation:
This module teaches how to save trained models to disk and load them later without retraining.

Learning Outcomes:

• Save model parameters
• Reload trained models
• Reuse models for inference or deployment

torch.save(model.state_dict(),"model.pth")
model.load_state_dict(torch.load("model.pth"))

Exercises

• Save model after training
• Reload model and predict

Assignment

Implement versioned model saving.

MCQ Quiz

• 1. state_dict stores?
  A. Model parameters
  B. Dataset
  C. Loss
  D. Code
• 2. torch.save() is used to?
  A. Save model
  B. Train model
  C. Evaluate
  D. Load data
• 3. Model loading requires?
  A. Same architecture
  B. Same data
  C. Same loss
  D. Same optimizer
• 4. Saved models are useful for?
  A. Deployment
  B. Data cleaning
  C. Visualization
  D. Preprocessing
• 5. File extension commonly used?
  A. .pth
  B. .csv
  C. .txt
  D. .json

Module 10: Binary Classification

Explanation:
This module integrates all previous concepts to build a complete binary classification model using a neural network.

Learning Outcomes:

• Build an end-to-end classification model
• Use sigmoid activation and BCELoss
• Train and evaluate a real-world model

import torch, torch.nn as nn

x = torch.randn(200,2)
y = (x.sum(1)>0).float().unsqueeze(1)

model = nn.Sequential(
    nn.Linear(2,8),
    nn.ReLU(),
    nn.Linear(8,1),
    nn.Sigmoid()
)

loss_fn = nn.BCELoss()
opt = torch.optim.Adam(model.parameters(), lr=0.01)

for _ in range(300):
    loss = loss_fn(model(x), y)
    opt.zero_grad()
    loss.backward()
    opt.step()

print("Training done")

Mini Project

• Generate your own dataset
• Train binary classifier
• Evaluate accuracy

Final Assignment

Build a spam vs ham classifier using PyTorch.

MCQ Quiz

• 1. Binary classification predicts?
  A. Two classes
  B. Multiple classes
  C. Numbers
  D. Images
• 2. Sigmoid output range?
  A. 0 to 1
  B. -1 to 1
  C. 0 to 10
  D. Any value
• 3. BCELoss is used for?
  A. Binary classification
  B. Regression
  C. Clustering
  D. Detection
• 4. Threshold commonly used?
  A. 0.5
  B. 1.0
  C. 0.1
  D. 0.9
• 5. This module combines?
  A. All previous concepts
  B. Only tensors
  C. Only datasets
  D. Only saving