Logging and Monitoring Programs Professional Debugging Workflows

Assertions and Defensive Programming Unit Testing with Pytest

(a.k.a. “How to Break Your Own Code Before the Customer Does”)

Congratulations — your ML program runs! Now it’s time for the real question:

“Will it still run tomorrow? On someone else’s computer? With real data?”

This chapter is where you become not just a coder, but a code detective 🕵️ — armed with pytest, logging, and a lot of caffeine. ☕

🧩 1. The Harsh Truth About Business Apps¶

In the business world, “it works” means absolutely nothing. Your model could predict stock prices correctly 95% of the time — but if it crashes during quarterly reporting, guess who’s getting the 2 a.m. Slack message? You. 🫠

Testing and debugging keep your application from turning into a “Friday afternoon deployment horror story.”

🧪 2. Testing: How to Catch Bugs Before They Catch You¶

Think of testing like a vaccine for your code — small, controlled failures to prevent catastrophic ones later.

🔹 Unit Testing¶

Tests the smallest building blocks — functions, classes, or modules.

Example (with pytest):

## test_discount_calculator.py
from discount_calculator import calculate_discount

def test_calculate_discount():
    assert calculate_discount(100, 0.1) == 90

Unit tests are like tiny bodyguards for your logic — they punch bugs in the face before they get inside.

Pro tip: Test edge cases, not just happy paths. If you only test when everything goes right, your app will collapse faster than a cheap IKEA shelf.

🔹 Integration Testing¶

Tests how your modules work together — because even if each one is perfect alone, they might fight when connected. ⚔️

def test_end_to_end_data_pipeline():
    data = load_data("sales.csv")
    transformed = clean_data(data)
    model = train_model(transformed)
    assert model.is_trained

Integration tests are like couple’s therapy for your code — making sure everyone gets along under stress.

🔹 Regression Testing¶

Business changes. Models get updated. Suddenly your new version makes negative revenue predictions. 😬

Regression tests ensure your new features didn’t break the old ones. (Like a “don’t embarrass me in front of the client” check.)

🔹 Automated Testing Pipelines¶

Tie it all together with CI/CD pipelines — because running tests manually is so 2010.

Example GitHub Action snippet:

- name: Run Unit Tests
  run: pytest --maxfail=1 --disable-warnings -q

If tests fail, your code doesn’t deploy. Yes, it’s strict — but so is gravity.

🐛 3. Debugging: Turning “Why?” Into “Ah, There It Is.”¶

Debugging isn’t a skill; it’s a lifestyle. It’s 40% logic, 60% emotional resilience.

Let’s break down the modern debugging toolkit:

🔹 Logging Like a Pro¶

Don’t print your way out of problems. Use the logging module — it’s the adult version of print().

import logging

logging.basicConfig(level=logging.INFO)
logging.info("Model training started")
logging.warning("Missing values detected in column 'price'")
logging.error("Model crashed because 'price' was a string")

Rule of thumb: If your log file looks like a bad breakup text thread, you’re doing it right. 💔📜

🔹 The Art of Reproducing Bugs¶

You can’t fix what you can’t repeat.

• Use seed control (np.random.seed(42)) for reproducible randomness.

• Snapshot data versions with DVC or MLflow.

• Log every environment variable that could betray you.

Debugging random bugs is like chasing a ghost that only appears when your manager walks in. 👻

🔹 The Power of the Debugger¶

Don’t underestimate IDE debuggers (like VS Code, PyCharm, or Jupyter’s %debug).

Set breakpoints, inspect variables mid-run, and say things like:

“Hmm, that list shouldn’t be empty…” until you achieve enlightenment. 🧘‍♂️

🔹 Mocking External Services¶

Your API shouldn’t actually hit a live payment system during tests. Use mocking to simulate external dependencies.

from unittest.mock import patch

@patch("payment_gateway.process_payment")
def test_checkout(mock_payment):
    mock_payment.return_value = True
    result = checkout(user="Alice", amount=50)
    assert result == "Success"

Mocking is pretending your system has friends, when really it’s testing alone in its room. 🤖

🧠 4. Testing Data and Models¶

Business apps aren’t just about code — your data can betray you too. Add data validation checks to your pipeline.

assert df["revenue"].notnull().all(), "Missing revenue values!"
assert (df["price"] > 0).all(), "Negative prices detected!"

Then, test your models for:

• Output range sanity (e.g., probabilities ∈ [0, 1])

• Drift detection (when today’s input looks nothing like last month’s)

• Bias testing (especially if your model makes business-critical decisions)

Your model isn’t wrong — it’s just “creatively confident.” 🎨

💻 5. Debugging Production¶

Production bugs are special. They don’t appear when you look for them. They only happen at 3 a.m., during a live demo, or right after the boss says,

“Looks stable now.” 😬

Survival Kit:¶

• Enable structured logging (JSON logs, timestamps, request IDs)

• Add exception alerts (via Slack, Datadog, Sentry)

• Use feature flags — turn off faulty features without redeploying

• Maintain a staging environment that mirrors production

If you’re debugging in production, congratulations — you’re now a full-stack firefighter. 🔥

⚖️ 6. When to Stop Debugging¶

Sometimes the problem isn’t your code — it’s the assumptions. When debugging takes more time than rewriting, walk away, breathe, and refactor.

Debugging burnout is real. Remember:

“A 3-hour bug fix is just a 5-line rewrite in disguise.” ✨

🏁 7. Final Thoughts: Test, Don’t Hope¶

Testing and debugging aren’t chores — they’re insurance policies for your sanity.

Golden rules:

• Test before you trust 🧪

• Log before you cry 📜

• Automate before you burn out 🤖

• Never say “it worked yesterday” again 🙃

“Good testing prevents bad surprises. Bad testing gives you character development.”

Exercises¶

Exercise¶