This project implements core Machine Learning algorithms from scratch using only Python and NumPy. No high-level ML libraries (scikit-learn, TensorFlow, PyTorch) are used.
- Hypothesis: ŷ = wx + b
- Loss function: Mean Squared Error (MSE)
- Optimization: Gradient Descent
- Error minimization
- Gradient descent mechanics
- Effect of learning rate
- Numerical instability and divergence
- Visualization of loss convergence
- Learning rate too high → divergence and NaN loss
- Learning rate too low → slow convergence
- Proper learning rate → smooth exponential decay
- Noisy data → approximation instead of perfect fit
The goal of this project is conceptual clarity, not performance. Each component is implemented step-by-step to understand how and why machine learning models learn.
- Python 3
- NumPy
- Matplotlib
Akhil