AIProject

A collection of machine learning algorithms built from scratch in Python, written as a hands-on way to understand the fundamentals before leaning on frameworks. Each algorithm is self-contained and interactive — inputs are prompted at runtime or read from a file.

---

Algorithms

Neural Network — algorithms/NeuralNetwork/NN.py

A fully connected feedforward network trained on the Iris dataset (binary classification — setosa vs. versicolor). Built with NumPy only. Supports configurable hidden layers, learning rate, and epoch count, all entered at runtime.

• Forward pass with ReLU activations and sigmoid output
• Backpropagation with gradient descent
• Sepal scatter plot shown before training for data visualization

K-Means Clustering — algorithms/clustering/clustering.py

An implementation of the K-Means algorithm on randomly generated 2D cluster data. Centroid positions are visualized before and after convergence.

• Random centroid initialization
• Iterative assignment and centroid update
• Tolerance-based convergence with a max iteration cap

Hidden Markov Model — algorithms/hiddenMarkov/hiddenMarkov.py

Runs the Viterbi algorithm over a randomly generated observation sequence. Reads state definitions from Input.txt — if the file is missing or empty, one is generated automatically.

Input format for Input.txt:

start: s0
s0:[0.7,0.3]:[0.4,0.6]
s1:[0.4,0.6]:[0.5,0.5]

Each line defines a state with its transition weights and emission probabilities.

Linear Regression — algorithms/linearReg/linearReg.py

Ordinary least squares regression on x,y data read from input.txt. If the file doesn't exist, random data is generated and saved. Plots the data and the best-fit line, then accepts a new X value for prediction.

Input format for input.txt:

1.0,2.3
2.5,4.1
3.8,6.0

---

Getting Started

Prerequisites

• Python 3.10+
• pip

Run

python run.py

This installs all required dependencies and presents a menu to select which algorithm to run. You can also run any script directly:

python algorithms/NeuralNetwork/NN.py
python algorithms/clustering/clustering.py
python algorithms/hiddenMarkov/hiddenMarkov.py
python algorithms/linearReg/linearReg.py

---

Dependencies

Package	Purpose
numpy	Core math and matrix operations
matplotlib	Plotting and visualization
pandas	Data formatting (Neural Network)
scikit-learn	Iris dataset and train/test split
torch / torchvision / torchaudio	Available for future use
seaborn	Available for future use

Full pinned versions are in requirements.txt.

---

Notes

• The neural network uses plain NumPy — no PyTorch in that module yet. It was written to build intuition for what frameworks do under the hood.
• The Hidden Markov Model auto-generates input if none is provided, so it works out of the box.
• Linear regression will prompt for point count and generate data if input.txt is missing.