Build With Me: Your First Predictor

Step 1: Setting up the Architecture

First, we define our hyper-parameters. These are the settings that control how fast our model learns and how long it trains.

import numpy as np

# Define how long to train, and how big of a step to take each time
EPOCHS = 100
LEARNING_RATE = 0.01

print(f"Initializing model for {EPOCHS} epochs...")

Step 2: The Training Loop

A model learns by making a guess, checking how wrong it is (Loss), and adjusting itself. Let's write that loop.

# Initialize random weights
m = np.random.randn() # Slope
b = np.random.randn() # Intercept

for epoch in range(EPOCHS):
    # 1. Make a prediction (y = mx + b)
    predictions = (m * X_train) + b
    
    # 2. Calculate the Error (Mean Squared Error)
    loss = np.mean((predictions - y_train) ** 2)
    
    # 3. Optimize (Simulated for this lab)
    # The model updates 'm' and 'b' to lower the loss
    
    if epoch % 10 == 0:
        print(f"Epoch {epoch} | Loss: {loss:.4f}")

Your Turn: Hit Train Model. Watch the visualizer on the left. The red line starts off terribly wrong. But as the epochs pass and the loss drops, watch how the line bends and snaps perfectly into the data points!

Step 3: Experimenting with Epochs

What happens if we don't train it long enough? Change EPOCHS = 100 to EPOCHS = 15 in your code, click Reset, and hit Train again.

You will see that the model stops training before the line fully fits the data. This is called Underfitting. To get good predictions, you must let the gradient descent find the optimal bottom of the loss curve.