ft_linear_regression

A linear regression implementation from scratch — predicts car prices from mileage using gradient descent, with no ML libraries involved.

---

How it works

The model fits a line price = θ₀ + θ₁ × km through the data. Training normalizes the inputs, runs gradient descent for 1000 epochs, then denormalizes the resulting parameters back to real-world scale.

Regression line

Loss over training

The loss drops fast in the first ~100 epochs and flattens out once the model has converged.

Precision

The trained model reaches an R² of 73.29% on the provided dataset.

---

Usage

1. Train

python3 src/train.py src/data.csv

Reads km and price columns from the CSV, runs gradient descent, and writes model.json.

2. Predict

python3 src/predict.py
# → Please enter a mileage: 80000
# → The estimated price of your car is : 6778

Loads model.json (falls back to θ₀=0, θ₁=0 if not found) and returns a price estimate.

3. Plot

python3 src/plot.py src/data.csv [model.json]

Opens an interactive scatter plot with the regression line overlaid.

4. Precision

python3 src/precision.py src/data.csv [model.json]

Prints the R² score of the model against the dataset.

---

Files

src/
├── train.py      — gradient descent, normalization, saves model.json
├── predict.py    — loads model, estimates price for a given mileage
├── plot.py       — scatter + regression line visualization
├── precision.py  — R² metric
└── data.csv      — 24 car records (km, price)

---

Install

pip install -r requirements.txt