How do we search a large configuration space without fooling ourselves?
Start here
How do we search a large configuration space without fooling ourselves?
HPO treats validation performance as an objective over configuration choices such as depth, learning rate or regularisation.
Building interactive view…
Understand
Build the mental model
HPO treats validation performance as an objective over configuration choices such as depth, learning rate or regularisation. Define a sensible search space and validation design first. More trials cannot rescue a leaky or noisy objective.
Click a stage to inspect what happens, what changes, and what should be checked before moving on.
Stage 1
Search space
For the “Search space” stage, identify the incoming object, the rule applied to it, the state change produced, and the evidence that would reveal a mistake. Technical context for Hyperparameter Optimisation: Grid search enumerates a lattice; random search samples dimensions efficiently; Bayesian/TPE methods use previous trials; Hyperband allocates budget adaptively.
Practitioner checkpoint: Define a sensible search space and validation design first. More trials cannot rescue a leaky or noisy objective.
What happens if…?
Break the assumption deliberately
Search a parameter that barely matters and see how grid search wastes budget compared with random search.
Move the control and explain what you expect before reading the visual.
Technical lens
Formalise what the visual is doing
Grid search enumerates a lattice; random search samples dimensions efficiently; Bayesian/TPE methods use previous trials; Hyperband allocates budget adaptively.
Technical questionUse a tiny case to make the mechanism observable. Grid search enumerates a lattice; random search samples dimensions efficiently; Bayesian/TPE methods use previous trials; Hyperband allocates budget adaptively. Verify one intermediate quantity, state change or mapping independently; then predict the consequence of this change: Search a parameter that barely matters and see how grid search wastes budget compared with random search.
Practitioner lens
Use it responsibly
Define a sensible search space and validation design first. More trials cannot rescue a leaky or noisy objective.
Transfer testTransfer this idea to a new example and justify each decision using this practitioner rule: Define a sensible search space and validation design first. More trials cannot rescue a leaky or noisy objective. Then explain what should change if you deliberately test: Search a parameter that barely matters and see how grid search wastes budget compared with random search.
Worked exploration
Use the visual as an experiment, not decoration
Define a fixed CV scheme and search space. Compare grid, random and adaptive search under the same evaluation budget. Keep the test set untouched while candidate configurations compete on validation evidence.
Technical lens
Grid search enumerates a lattice; random search samples dimensions efficiently; Bayesian/TPE methods use previous trials; Hyperband allocates budget adaptively.
Practitioner check
Define a sensible search space and validation design first. More trials cannot rescue a leaky or noisy objective.
Prediction before interaction
Search a parameter that barely matters and see how grid search wastes budget compared with random search.
Exploration walkthrough
Turn the interaction into an evidence trail
Define a fixed CV scheme and search space. Compare grid, random and adaptive search under the same evaluation budget. Keep the test set untouched while candidate configurations compete on validation evidence. Before moving the control, state your prediction. After the visual changes, name the specific state, statistic, boundary or mapping that changed and explain why that change is consistent—or inconsistent—with your prediction.
Record one observable quantity before the interaction and the same quantity afterwards.
Change one factor at a time so the causal effect of the control is inspectable.
Use an edge or failure case to discover where the concept stops behaving as the simple story suggests.
Static orientation diagram for Hyperparameter Optimisation; use the interactive visual above to test how the relationships change.
Reference depth
Open the complete material
The flagship experience is the map. These pages contain the roads.
