How 0bada1/ft_linear_regression works
Project Type and Purpose
• This is a command-line machine-learning utility (a single-feature linear regression tool) whose goal is to train a model that predicts a car’s price from its mileage, generate visualizations of cost and parameter convergence, then offer a prediction interface and precision evaluation.
High-Level Components
a. Data Layer
• Raw data files (data/data.csv, data/test_data.csv)
• CSV loader (srcs/train_model/load_data.py)
b. Training Pipeline
• Cost computation (cost.py)
• Gradient descent optimizer (gradient_descent.py)
• Model definition and utility functions (linear_regression.py, plot.py, log/)
• Orchestrator script (main.py & train_model.sh)
• Output: trained parameters (model.txt) and plots (PNG or stdout)
c. Prediction Module
• CLI wrapper (predict.py)
• Loads model.txt, accepts mileage input, outputs predicted price
d. Precision Evaluation Module
• calculate_precision.py
• Loads test_data.csv and model.txt, computes prediction accuracy metrics
e. External Libraries & Tools
• Python 3.x, NumPy, Matplotlib (for plotting), shell script for orchestration
Relationships & Data Flow
• Data Layer → Training Pipeline: raw CSV → load_data.py → cost & gradient computation → model.txt & plots
• Training Pipeline → Prediction Module: model.txt consumed by predict.py
• Training Pipeline & Data Layer → Precision Module: test_data.csv + model.txt → calculate_precision.py
• Optionally, Training Pipeline logs intermediate theta values in log/ for plotting module
Architectural Patterns & Principles
• Pipeline pattern: Clear stages (Load → Train → Evaluate → Predict)
• Separation of concerns: each functionality in its own directory (train_model, predict, precision)
• Single responsibility: cost, gradient, data loading, plotting each isolated into its own module
Essential Elements
• Clear, concise component names (e.g., “Training Pipeline”)