This repository contains course notes and materials for MSDS621, an introductory machine learning course at the University of San Francisco. It is designed for students seeking a deep understanding of core ML models and processes, with a focus on practical implementation and model interpretation. The benefit is a hands-on approach to learning key algorithms through Python coding.

How It Works

The course emphasizes deep dives into a few key models rather than a broad survey. Students implement algorithms like regularized linear regression, Naive Bayes, decision trees, and random forests from scratch using Python. This hands-on approach, inspired by Richard Feynman's philosophy, aims to build intuition and understanding of model behavior, data cleaning, feature engineering, and model assessment.

Quick Start & Requirements

• Primary Install: Python 3
• Prerequisites: Google Chrome and a webcam are required for exams via HonorLock. Access to a GPU (via Colab or AWS) is recommended for specific GPU-accelerated MNIST examples.
• Resources: Links to course notebooks and slides are provided within the README.

Highlighted Details

• Focus on implementing models from scratch (linear regression, Naive Bayes, decision trees, random forests).
• Covers regularization (L1, L2) and gradient descent for training.
• Includes modules on model interpretation (feature importance, partial dependence) and unsupervised learning (clustering).
• Introduces vanilla neural networks using PyTorch, including autograd and GPU acceleration.

Maintenance & Community

• Instructor: Terence Parr.
• Communication: Slack is used for course-related communication.

Licensing & Compatibility

• Licensing is not explicitly stated in the README.
• Materials are intended for educational use within the MSDS621 course.

Limitations & Caveats

The course materials are specific to the Fall 2021 semester and may not reflect current best practices or updated library versions. Exams require HonorLock, which has strict proctoring and technical requirements. The grading policy is binary with no partial credit for projects.