AI From Scratch

This repository contains implementations of various AI and machine learning concepts, architectures, and exercises built from scratch. It serves as a learning resource for understanding the underlying principles of artificial intelligence and machine learning algorithms.

Repository Structure

Architectures

This directory contains implementations of different neural network architectures and machine learning models.

• convolutional_neural_network_reinforcement_learning_monte_carlo_tree_search_selfplay_alphazero_tictactoe.ipynb: Implementation of a convolutional neural network with reinforcement learning using Monte Carlo Tree Search and self-play, similar to AlphaZero, applied to Tic-Tac-Toe.
• logistic_regression.ipynb: Implementation of logistic regression algorithm.
• lstm.py: Long Short-Term Memory neural network implementation.
• diffusion.py: Diffusion model implementation.
• linear_regression.py: Linear regression implementation.
• bigram.py: Bigram language model implementation.
• multiple_linear_polynomical_sinus_etc_regression_and_gradient_descent.ipynb: Implementation of various regression techniques including multiple linear, polynomial, and sine regression with gradient descent.
• reinforcement-learning-deep-q-learning: Implementation of Deep Q-learning reinforcement learning and applied to Snake game.
• grpo_group_relative_policy_optimization.ipynb: Implementation of Group Relative Policy Optimization (GRPO), a reinforcement learning algorithm that enhances LLM's reasoning abilities by generating multiple responses to a given prompt, evaluating each using a reward function (solution correctness in math), and updating the model based on the relative performance of these responses within the group

Transformer

• transformer.py: Implementation of the transformer architecture.
• einops.py: Implementation of einops operations for tensor manipulations.

GPT-2

• train_gpttwo.py: Training script for a GPT-2 style model.
• play.ipynb: Interactive notebook for playing with the trained GPT-2 model.
• fineweb.py: Fine-tuning utilities for web data.
• hellaswag.py: Implementation for the HellaSwag benchmark.
• input.txt: Training data for the model.

Physics-Inspired Neural Networks

• cooling/: A collection of notebooks and code exploring physics-based applications of machine learning:
  • temp_pred.ipynb: Neural network models and physics informed neural network model for predicting temperature dynamics in cooling systems, including implementation of L2 regularization techniques.
  • regularisation_ex.ipynb: Demonstrates the application of regularization techniques in machine learning models to prevent overfitting, with visualizations comparing regularized vs. non-regularized polynomial regression.
  • network.py: Basic neural network architecture for solving physics-based problems.
  • diff_equations.py: Implementation of fundamental cooling law equations and gradient calculation for physics-based machine learning.

Exercises

• hyperplane_classifier_of_clothes.py: Exercise implementation of a hyperplane classifier for clothing items.

Getting Started

To use this repository, clone it to your local machine and explore the different implementations. Each file is self-contained and includes the necessary code to understand and run the respective algorithm or model.

git clone https://github.com/yourusername/ai-from-scratch.git
cd ai-from-scratch

Prerequisites

• Python 3.x
• NumPy
• PyTorch (for some implementations)
• Jupyter Notebook (for running .ipynb files)

License

This project is available for educational purposes.

Acknowledgments

• Inspired by the desire to understand AI and machine learning concepts from first principles.