Chapter 2: Testing ML Systems

Types of ML Tests

Test Type	What It Tests	Example
Unit	Individual functions	Feature engineering
Data	Data quality	Schema, null values
Model	Model behavior	Predictions on known inputs
Integration	End-to-end	API to prediction
Performance	Speed/resource usage	Latency, memory

Example Tests

# tests/test_model.py
import pytest
import numpy as np

class TestModel:
    @pytest.fixture
    def model(self):
        return load_model("models/churn_model.pkl")
    
    def test_prediction_shape(self, model):
        X = np.random.randn(10, 15)
        predictions = model.predict(X)
        assert predictions.shape == (10,)
    
    def test_prediction_range(self, model):
        X = np.random.randn(100, 15)
        predictions = model.predict_proba(X)
        assert np.all(predictions >= 0) and np.all(predictions <= 1)
    
    def test_known_input(self, model):
        # High-risk customer features
        high_risk = np.array([[0, 100, 365, 0, 1, ...]])
        prob = model.predict_proba(high_risk)[0, 1]
        assert prob > 0.7, "High-risk customer should have high churn probability"

---

Next: Chapter 3: Continuous Training