Simple Neural Network Example

import math
import random

random.seed(42)

# ====================================================
# Simple 2-layer neural network for binary classification
# Task: predict high (1) or low (0) credit risk
# Features: [income_scaled, debt_ratio]
# ====================================================

# Training data: [income_scaled, debt_ratio] -> credit_risk
data = [
    ([0.8, 0.2], 0),  # high income, low debt -> low risk
    ([0.3, 0.8], 1),  # low income, high debt -> high risk
    ([0.9, 0.1], 0),  # very high income, minimal debt -> low risk
    ([0.2, 0.9], 1),  # very low income, very high debt -> high risk
    ([0.6, 0.4], 0),  # decent income, moderate debt -> low risk
    ([0.1, 0.7], 1),  # low income, high debt -> high risk
]

# Helper functions
def sigmoid(z):
    return 1 / (1 + math.exp(-z))

def predict(x1, x2, w1, w2, bias):
    """Simple 1-neuron model (no hidden layer for clarity)."""
    z = x1 * w1 + x2 * w2 + bias
    return sigmoid(z)

# Initialize weights randomly
w1 = random.uniform(-1, 1)
w2 = random.uniform(-1, 1)
b  = 0.0
lr = 0.5

print("Training a neural network on credit risk data...")
print()

# Train for 200 epochs
for epoch in range(200):
    total_loss = 0
    for features, label in data:
        x1, x2 = features
        
        # Forward pass
        output = predict(x1, x2, w1, w2, b)
        
        # Loss (binary cross-entropy, simplified)
        loss = -(label * math.log(output + 1e-8) + (1 - label) * math.log(1 - output + 1e-8))
        total_loss += loss
        
        # Gradient descent update
        error = output - label
        w1 -= lr * error * x1
        w2 -= lr * error * x2
        b  -= lr * error

    if epoch % 50 == 0:
        accuracy = sum(
            1 for f, l in data if round(predict(f[0], f[1], w1, w2, b)) == l
        ) / len(data)
        print(f"Epoch {epoch:3d}: loss={total_loss:.3f}, accuracy={accuracy*100:.0f}%")

print()
print("Final predictions on training data:")
for features, label in data:
    pred = predict(features[0], features[1], w1, w2, b)
    predicted_class = "HIGH RISK" if round(pred) == 1 else "LOW RISK"
    actual_class = "HIGH RISK" if label == 1 else "LOW RISK"
    status = "OK" if predicted_class == actual_class else "WRONG"
    print(f"  Income={features[0]}, Debt={features[1]} -> {predicted_class} (actual: {actual_class}) {status}")