Start from scratch

AI Foundations for a Beginner

Build, test, and explain a small classifier while recognizing leakage, overfitting, and misleading metrics.

10-13 hours · estimatedSelf-guided · learn at your pace
Before you start

Learners who can use a spreadsheet and are willing to write a little Python.

  • Basic algebra
  • Ability to install Python or use a free browser notebook
  • No prior machine-learning experience
LESSON 1 · 2-2.5 hours ESTIMATED

From a question to a learning problem

Distinguish prediction, classification, clustering, and generation, then define one supervised-learning task precisely.

Machine learning fits a mathematical rule from examples. The first skill is deciding what each example contains, what answer is available during training, and what decision the output will support. A house-price target is numeric, so it suggests regression; a flower species is categorical, so it suggests classification. Clustering has no supplied answer and can reveal groups, but the learner must interpret those groups. Generative models create new content from learned patterns and need different evaluation. For a small first experiment, load scikit-learn’s Iris dataset and write a one-sentence task specification: inputs, target, unit of observation, and intended use. Inspect the shape, feature names, class counts, and five rows. Then make a deliberately simple rule using petal length and record its errors. This separates the scientific question from the software. A model is useful only relative to a defined task, population, and error cost; high accuracy without those definitions says little.

Try it yourself

  1. Install Python, pandas, matplotlib, and scikit-learn, or open a free local Jupyter environment.
  2. Load sklearn.datasets.load_iris and inspect dimensions, fields, missing values, and class balance.
  3. Write and test one hand-built threshold rule before fitting a model.

You’re ready to move on when…

  • Task statement names inputs, target, observation unit, and intended use.
  • Notebook reports the dataset dimensions and class counts correctly.
  • A confusion table exposes at least one error made by the hand-built rule.
Enroll above to keep your progress across visits.
PUT IT INTO PRACTICE

A reproducible Iris classifier report

Build a Pipeline-based classifier, preserve a final stratified test set, compare two models, choose a metric for a stated use case, and publish a notebook plus one-page model card.

What you’ll make

  • Runnable notebook
  • Confusion matrix and threshold plot
  • Model card
  • requirements.txt or environment file

Check your work

  • Fresh execution completes without manual edits.
  • Test data never enters fitting or model selection.
  • Report includes a chance or simple-rule baseline and at least three limitations.
Project estimate: 2-3 hours. Completion is self-reported; this is not an accredited qualification.