LearningMachines

A collaborative, no-frameworks (beyond NumPy) machine learning repo built from first principles.
We implement core algorithms line-by-line to understand the math, not just call .fit().

Status: ✅ Logistic Regression (from scratch) · ⏭️ Next: Support Vector Machines

---

Table of Contents

• Goals
• What’s Inside
• ToDo
• Roadmap

---

Goals

• Demystify ML by implementing algorithms from scratch.
• Build a friendly collaboration space for anyone and everyone.
• Keep code minimal, readable, and well-documented.
• Add tests and examples as we go.

---

What’s Inside

• Linear Regression (from scratch)

  • Gradient Descent optimization
  • Stochastic Gradient Descent optimization
  • Arbitrary feature dimensions
  • Minimal dependencies

• Logistic Regression (from scratch)

  • Gradient Descent optimization
  • Minimal dependencies
  • Logistic hypothesis (sigmoid)
  • Cross-entropy cost
  • Gradient Descent
  • Weight initialization
  • Feature standardization
  • Proper train–validation split
  • Convergence checks
  • Forward + backward pass
  • Evaluation pipeline (MSE/RMSE)
  • End-to-end NumPy workflow

ToDo

ToDo : Add docstrings and comments in model.py [Linear Regression and Logistic Regression from scratch]

---

Roadmap

• Linear Regression (GD & SGD)
• Logistic Regression (Binary)
• Support Vector Machines
• Metrics: MSE, MAE, Accuracy, Precision/Recall
• Train/Val split utilities
• Visualization helpers
• Unit tests (pytest)
• Additional models: KNN, Decision Trees