Simple Neuron

simple-neuron.py

# We set a fixed starting point for random numbers so we get the same results each time
import random
random.seed(42)

# Our learning data
x_data = [1, 2, 3, 4, 5]                # Input numbers
y_data = [2, 4, 6, 8, 10]               # Output numbers (each is 2 times the input)

# Starting guesses for our formula (y = w*x + b)
w = random.random()                     # This should end up close to 2
b = random.random()                     # This should end up close to 0

# How quickly to adjust our guesses (small steps)
lr = 0.01

# Try to get better 1000 times
for epoch in range(1000):
    total_loss = 0
    for x, y in zip(x_data, y_data):
        # Make a guess
        y_pred = w * x + b              # prediction
        
        # Check how wrong we are
        loss = (y_pred - y) ** 2        # Square the error to make it positive
        total_loss += loss              # Add to total mistakes

        # Figure out how to improve
        grad_w = 2 * (y_pred - y) * x   # How to change w
        grad_b = 2 * (y_pred - y)       # How to change b

        # Make small improvements
        w -= lr * grad_w                # Update w
        b -= lr * grad_b                # Update b

    # Show progress every 100 tries
    if epoch % 100 == 0:
        print(f"Epoch {epoch}: Loss = {total_loss:.4f}, w = {w:.4f}, b = {b:.4f}")

# Show what we learned
print(f"Trained weight: {w:.4f}, bias: {b:.4f}")
# If it worked, w should be close to 2 and b close to 0