PyTorch Linear Regression Project

This project demonstrates a simple linear regression model implemented using PyTorch. It covers the entire machine learning workflow, including data generation, model creation, training, evaluation, and saving/loading the model.

Data Generation

The project creates synthetic data using the equation y = wx + b, where:

• w (weight) = 0.7

• b (bias) = 0.3

• It produces 50 datapoints in which 40 data points are used for training and 10 for testing

Model Architecture

A custom LinearRegressionModel class is defined, inheriting from nn.Module. It includes:

• Learnable parameters: weight (w) and bias (b)
• A forward method defining the linear regression equation

Training

The model is trained using:

• Loss Function: L1Loss (Mean Absolute Error)
• Optimizer: Stochastic Gradient Descent (SGD)
• Number of epochs: 200

Visualization

The project includes functions to visualize:

• Training and testing data
• Model predictions
• Loss curves during training

Results

loss curves

*

prediction before training

prediction after training