brightlab.Made for curious mindsBrightlab
Year 2 · Training data
Where will you look for creatures? What might you miss?
Go to the investigation ↓A before/after habitat coverage map and a justified collection plan.
Where will you look for creatures? What might you miss?
Examples help only with what they show. A collector who stays in the middle of a habitat can gather many copies and still miss creatures at the edges. Our coverage model checks whether each test creature has a nearby training example. Coverage is not the same as accuracy: being near an example does not prove its label is correct.
Before this lesson: Count examples inside and outside a drawn boundary. Useful earlier investigations: The sorting stream.
Examples help only with what they show. A collector who stays in the middle of a habitat can gather many copies and still miss creatures at the edges. Our coverage model checks whether each test creature has a nearby training example. Coverage is not the same as accuracy: being near an example does not prove its label is correct.
Count examples inside and outside a drawn boundary. Useful earlier investigations: The sorting stream.
Scatter 12 paper creatures across a habitat strip, including two at each end. Cut a movable collection window. Rehearse the fixed test positions and explain that the coverage radius is a teaching choice.
ACARA V9 AI curriculum connection ↗ · Technologies ↗
A complete teaching sequence · 45 minutes
Take several copies of the central creature. Count them beside a smaller varied collection.
Ask: “Which collection tells us about the edge?”
Listen for: “The one with creatures from different places.”
Predict how copying central examples changes the number of covered test creatures.
Ask: “Will an extra copy move an example closer to the edge?”
Listen for: “No, it stays in the middle.”
Change window width with the habitat and coverage radius fixed. Inspect covered test positions and gaps; record counts.
Ask: “Which test creature became covered first?”
Listen for: “The one just outside the old window.”
Increase duplicate copies without widening the window and compare coverage.
Ask: “The collection grew. What did not improve?”
Listen for: “The uncovered edges stayed uncovered.”
Choose a wider or shifted window that covers both extremes. Draw a sampling plan and identify one remaining gap.
Ask: “How would you collect differently next time?”
Listen for: “Go to the places we have not sampled.”
Present count and coverage as separate measures.
Ask: “Does covered mean certain to be right?”
Listen for: “No, the nearby example may still have the wrong label.”
More copies mean better learning.
Many duplicates increase sample count but leave the nearest-example distances at habitat edges unchanged.
Improve coverage by changing where examples are collected while keeping all test positions fixed.
Check that a student can point to a gap and distinguish a larger count from a wider spread. Ask them to predict one new collection position.
| Criterion | Beginning | Secure | Extending |
|---|---|---|---|
| Sampling reasoning | Equates more with better | Distinguishes duplicates from variety | Explains a specific coverage gap |
| Collection redesign | Adds arbitrary copies | Adds examples near uncovered tests | Justifies coverage gains and remaining limitations |
Use six large test positions and coloured strings to show a nearby-example radius.
Fix the collection budget at four examples; spread them to minimise the largest uncovered gap.
All data are synthetic. A real habitat survey would require teacher supervision and environmental permissions; the browser requires no field collection.
Shared device? Turn remembering off. A project file lets you continue on another device.
Your browser is the laboratory
Where will you look for creatures? What might you miss?
Twelve test positions span habitat 0.5–9.5. The initial collection window is centred at 5 with width 3. A test is covered if a training example is within 1 unit. Copies stay at the same position.
The experiment opens after your prediction.
There is no penalty for being surprised.
Apple Silicon · PyTorch MPS
Compare thousands of random and stratified sampling plans against a fixed synthetic habitat; display coverage distributions rather than a single lucky sample.
Teacher-operated extension: students predict and interpret the visual report together.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download Mac Studio investigation ↓unzip y2-coverage-mac-pathway.zip -d y2-coverage-mac cd y2-coverage-mac bash setup-mac.sh source .venv/bin/activate python experiment.py --device mps --output results marimo edit notebook.py
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
NVIDIA DGX Spark · PyTorch CUDA
Compare thousands of random and stratified sampling plans against a fixed synthetic habitat; display coverage distributions rather than a single lucky sample.
Teacher-operated extension: students predict and interpret the visual report together.
The bundle contains lesson-specific working code, a configuration file, a reactive notebook, a deterministic CPU check and hardware setup instructions. Acceleration is reported only after a tensor operation and result read-back succeed.
Download DGX Spark investigation ↓unzip y2-coverage-dgx-pathway.zip -d y2-coverage-dgx cd y2-coverage-dgx bash run-dgx.sh # Open the localhost notebook URL printed by the container.
CPU and available-device execution status is recorded in the downloaded README and validation report. DGX Spark execution requires that hardware; static validation alone does not establish GPU compatibility or performance. The browser lesson remains fully available without this extension.
A before/after habitat coverage map and a justified collection plan.
Download the editable handout →