How This Course Works
How Machine Learning with Python works on ask-codey.com, from reading a lesson and trying its Do It to exercises with their own data, a notebook to explore and a module pytest checks, chapter checks, module assessments, the projects you build, and why every number in a lesson can be reproduced.
Novice · AI Basics · Python · Updated
After this lesson you can
- Describe how this machine learning course's lessons, Do Its, Check yourself blocks, exercises, chapter checks, assessments and projects fit together
Key takeaways
- The course has five modules. Each one ends with a mini-capstone, a project of its own, and an assessment that passes at 80%, with three tries a day.
- Skills are short "you can do this" statements. They are what the questions test and what your progress counts.
- An exercise comes with its own data, a notebook to explore in and a module to finish. pytest checks the module, never a score that depends on luck.
- Every number in a lesson can be reproduced, because the data is a saved file, the package versions are pinned and every random step gets a fixed seed.
- Covered counts the lessons you've finished. Mastered counts the skills you've shown you know, with three clean answers from at least two different angles.
This course is self-study: you go at your own pace, and the course itself tells you how you're doing. Here is what it's made of, and how to get the most from each part.
The shape of the course
| Part | What it is |
|---|---|
| Module | A big stage of the course, ending with a project |
| Chapter | One subject, such as Getting Started with Machine Learning or Measuring a Classifier |
| Topic | A group of lessons inside a chapter |
| Lesson | One idea, explained with small examples you can run |
The five modules, and the project each one ends with:
| Module | What you learn | The project |
|---|---|---|
| 1. Learning from Examples | Setup, what a model is, training and testing, a little statistics | Is Machine Learning the Right Tool? |
| 2. Predicting Numbers | First models with scikit-learn, and how far off they are | Bike Share Forecast |
| 3. Predicting Categories | Models that sort into categories, and their mistakes | Library Renewals |
| 4. Making Models Better, Honestly | Overfitting, cross-validation, pipelines and leakage | The Model That Only Seemed to Work |
| 5. Groups, Fairness and Explaining | Groups without labels, explaining and fairness | Shopper Groups |
Each lesson has a level from 1 to 4: level 1 meets an idea for the first time, and higher levels go deeper.
Skills, Do Its and Check yourself
Most lessons list their skills: short "you can do this" statements, such as Explain the difference between a rule you write by hand and a rule a model learns from examples. Skills are what the questions test and what your progress counts.
Some lessons have a Do it: a small task in the lesson's own example, solved in a box right on the page, with tests that tick as your code works. Python runs in your browser for it, with its standard library only, so lessons whose idea works in plain Python have one; lessons that need pandas or scikit-learn are practiced in the chapter's exercise, on your own computer.
A Check yourself block near a lesson's end asks one to three short questions, then shows the answer and why. A wrong answer is useful news, not a failure: read the explanation, look back at the lesson, and carry on. Each lesson ends with its Sources: the documentation pages its facts were checked against.
Exercises
Exercise: lessons give you a small piece of machine learning work to finish, as a folder of starter files on GitHub: its own data (a CSV file of our own made-up data, with the seed_data.py script that made it), a notebook to explore in, a module to finish (a .py file whose functions do the work), and tests, small programs that call your functions and check what comes back.
You copy it with Git, open it in VS Code, make its virtual environment (its README shows how), and run the tests in the terminal:
python -m pytest
pytest is a test runner: it finds the tests in the folder, runs them, and lists which pass and which fail. At first they fail, and that's expected: each function you finish turns its tests green. The tests check the module, because a test can call a function in a .py file but not a notebook cell, and they never check a score that depends on chance: they check what you built, such as a split's sizes, and arithmetic with one right answer.
Numbers you can reproduce. Machine learning uses chance on purpose, such as shuffling rows before a split. So that your numbers match the lesson's, everyone uses the saved CSV, every package is pinned to an exact version in requirements.txt, and every step that shuffles gets a random seed: a fixed starting number, such as random_state=42, so it shuffles the same way every time.
When every test passes, you push your work to a repository of your own on GitHub and submit its address on the lesson page, which completes the lesson. Get the Exercise Code with Git and Hand In Your Work with Git show each step.
Checks, assessments and projects
- A chapter check is a short quiz on the chapter's skills, with a few from the chapter before. Take it as often as you like: it's practice, and it tells you what to review.
- A module assessment ends each module. 80% passes. You get three tries a day, then a 24-hour rest before the next three. Every attempt draws different questions, so a retake is a fair new try, not a memory test.
- A mini-capstone ends each module: a project using everything the module taught, with a brief, starter files, milestones that break it into steps, and a short checklist, called a rubric, that you score your own work against.
- The final capstone is the course's final assessment; there's no big final test. You build a model for a question you choose, compare and measure it honestly, and explain it.
Covered, Mastered, and who gets what
Signed in, the course keeps two separate records of your progress:
- Covered: the lessons you've finished.
- Mastered: the skills you've shown you know. A skill is mastered when you answer three different questions about it cleanly, meaning right the first time and without a hint, from at least two different angles: one might ask what a split holds, another which line splits it. A later miss drops it back, so mastery stays honest.
Everyone can read every lesson and answer its questions. Signed in, your answers count toward your skills, and you can tick lessons as done. With Premium, or if you own this course, you also get course progress on every page, the chapter checks, the module assessments, and the Certificate of Completion: every lesson covered and every module assessment passed.
A good rhythm: read a lesson and run its examples, answer Check yourself, do the chapter's exercise and its check, and when a module's skills are mastered, take its assessment and build its project.
Check yourself
1 question to try on the lesson page.
So what
The course is built so you always know where you stand, and so every number you see is one you can get yourself: what you've read, what you've mastered, what to practice next, and what you've built.