Machine Learning Lab
Vélanámsstofan
Train your own models. Trust the numbers, not the hype.
What they'll learn
- Work real datasets Load, clean and explore real data the way data scientists do.
- Train models that predict Build classification and regression models that make real predictions.
- Evaluate honestly Learn where accuracy deceives and how to judge a model truthfully.
- Find the bias Spot bias in data and models, and explain it clearly.
A trained model on a dataset they chose, with an honest model card.
The 13-week journey
-
Sprint 1 — Data first
Notebooks, pandas-light: loading, cleaning and charting real data; what a feature and a label are; train/test split. Ends with a charted dataset they explored.
-
Sprint 2 — Train & evaluate
Classification and regression with scikit-learn; accuracy, and the metrics-honesty lab where accuracy lies; comparing models fairly. Ends with a defensible trained model.
-
Sprint 3 — Fair & honest
Bias probes: train narrow, test wide; cleaning data to improve a model; writing a model card that states the limits. Choosing a dataset that matters to them.
-
Week 13 — Showcase
Each student presents their model, its data story, and what it can't do — model cards on the wall.
What we cover
Every topic, unit by unit — so you know exactly what your child builds and learns.
Data first
- Notebooks and pandas-light
- Loading, cleaning and charting real data
- Features and labels
- The train/test split
Train & evaluate
- Classification and regression with scikit-learn
- Accuracy and other metrics
- The metrics-honesty lab — where accuracy lies
- Comparing models fairly
Fair & honest
- Bias probes: train narrow, test wide
- Cleaning data to improve a model
- Writing a model card with honest limits
- Choosing a dataset that matters
Showcase & model cards
- Telling the data story
- Presenting what the model can't do
- Model cards on the wall
A defensible trained model with its data story and accuracy card, presented like a young data scientist.
The technical elective for teens who want to go under the hood: load real data, train your own models, and learn to judge them like a scientist. It turns the magic of AI Makers into method, and sets up the serious 16–18 track.
The hooks
Kid hook: “I trained my own model — and I can prove how good it really is.” Parent hook: “Real data-science skills with honest evaluation — the substance behind the buzzword.”
Who it’s for
13–15s ready for the technical side; some Python (AI Makers or equivalent) helps. Thrives: the analytical, the curious, the future engineers.
Outcomes — by the end, students can
Load, clean and chart a dataset; train classification and regression models; choose metrics that don’t mislead; probe for bias and improve with better data; write an honest model card.
Tools & compliance
Python notebooks, pandas and scikit-learn on school accounts; public/Icelandic open datasets; 13+ accounts with signed parental consent; camera-free.
Where this course fits
The bridge from AI Makers into the technical 16–18 track (AI Engineer, Applied ML).
Parent questions
Is this too advanced for 13–15?
No — we build intuition carefully and use friendly tools; it's hands-on, not lecture-heavy. Some Python (e.g. from AI Makers) helps.
What math is needed?
School math is enough; we teach the ideas through building, never hiding behind notation.
What do they leave with?
A trained model they can explain, and an honest model card — real data-science habits.
The first lesson is a free trial.
Book a no-commitment trial — pay nothing if it's not a fit.
Book a free trial