ml-from-scratch / src /linear_regression.py
Eric-Yyvuw's picture
Upload src/linear_regression.py with huggingface_hub
8e00c30 verified
Raw History Blame Contribute Delete
3.05 kB
"""
Linear Regression with Gradient Descent — CS401
From scratch, no ML libraries.
"""
import random
# ---- Training Data ----
# Living area (m²) → Price (€k)
X = [50, 60, 70, 80, 90, 100, 110, 120, 130, 140]
Y = [150, 180, 210, 240, 270, 300, 330, 360, 390, 420]
n = len(X)
# ---- Hypothesis ----
def predict(x, theta0, theta1):
"""h(x) = theta0 + theta1 * x"""
return theta0 + theta1 * x
# ---- Loss Function (MSE) ----
def compute_loss(theta0, theta1):
"""J(theta0, theta1) = (1/n) * sum((h(x_i) - y_i)^2)"""
total = 0
for i in range(n):
y_pred = predict(X[i], theta0, theta1)
total += (y_pred - Y[i]) ** 2
return total / n
# ---- Evaluation Metrics ----
def mae(theta0, theta1):
"""Mean Absolute Error = (1/n) * sum(|h(x_i) - y_i|)"""
total = 0
for i in range(n):
total += abs(predict(X[i], theta0, theta1) - Y[i])
return total / n
def rmse(theta0, theta1):
"""Root Mean Squared Error = sqrt((1/n) * sum((h(x_i) - y_i)^2))"""
return compute_loss(theta0, theta1) ** 0.5
# ---- Gradient Descent ----
def gradient_descent(lr=0.00001, iterations=1000):
theta0 = 0 # bias
theta1 = 0 # weight
for iteration in range(iterations):
# Compute gradients
grad0 = 0 # dJ/dtheta0
grad1 = 0 # dJ/dtheta1
for i in range(n):
y_pred = predict(X[i], theta0, theta1)
error = y_pred - Y[i]
grad0 += error # derivative w.r.t. theta0
grad1 += error * X[i] # derivative w.r.t. theta1
grad0 /= n
grad1 /= n
# Update parameters
theta0 -= lr * grad0
theta1 -= lr * grad1
# Print progress every 200 iterations
if iteration % 200 == 0:
loss = compute_loss(theta0, theta1)
print(f"Iter {iteration:4d} | Loss: {loss:>10.2f} | theta0: {theta0:.4f} | theta1: {theta1:.6f}")
return theta0, theta1
# ---- Main ----
if __name__ == "__main__":
print("=" * 60)
print("Linear Regression — Gradient Descent")
print("=" * 60)
print(f"Training data: {n} examples")
print(f"X (area m²): {X}")
print(f"Y (price €k): {Y}")
print()
theta0, theta1 = gradient_descent(lr=0.00001, iterations=2000)
print()
print("=" * 60)
print(f"Result: h(x) = {theta0:.2f} + {theta1:.4f} * x")
print(f"Final loss (MSE): {compute_loss(theta0, theta1):.2f}")
print(f"MAE: {mae(theta0, theta1):.4f}")
print(f"RMSE: {rmse(theta0, theta1):.4f}")
print("=" * 60)
# Predictions
print()
print("Predictions:")
print(f"{'Area (m²)':>10} | {'Actual (€k)':>12} | {'Predicted (€k)':>14}")
print("-" * 42)
for x_val, y_val in zip(X, Y):
y_pred = predict(x_val, theta0, theta1)
print(f"{x_val:>10} | {y_val:>12} | {y_pred:>14.2f}")
# Predict new value
print()
new_area = 150
predicted_price = predict(new_area, theta0, theta1)
print(f"New prediction: {new_area} m² → {predicted_price:.2f} €k")