Experiment Design Best Practices
A well-designed experiment optimizes faster and produces more useful results. This checklist summarizes the most important decisions to get right before launching an optimization.
Quick Checklist
Area | Recommendation |
Parameters | Keep the count low — fix anything that does not matter. Add properties for categorical variables. |
Objectives | Use a scalar strategy (Weighted Sum or Hierarchy) when tradeoffs are known. Reserve Pareto for exploration. Use Hierarchy to include cost as a secondary objective. Track non-linear derived quantities as results without objective. |
Constraints | |
Expert context | Set during experiment setup only (fixed once started). Influences initial design (except for constrained experiments) and insights throughout. Focus on optimality regions, objective ranges, and similarity to known problems. |
Optimization Difficulty Rating
As you configure your experiment, SDLabs evaluates the complexity of your optimization problem and displays a difficulty rating — a color-coded badge that updates live as you edit.
Rating | Meaning |
Optimal (green) | The configuration is well-suited for optimization. No issues detected. |
Moderate (yellow) | Some factors may slow convergence. Review the warnings and adjust if possible. |
Challenging (red) | Multiple complexity factors detected. The optimizer will still work, but convergence may require more iterations. |
Expand the rating badge to see specific warnings — each one explains what is adding complexity and suggests how to simplify. Common factors include:
High dimensionality — more than 20 parameters
Mixed variable space — combining continuous and discrete variables
Large search space — too many discrete combinations
Many objectives — more than 3 for Pareto, or more than 6 for scalar strategies
The rating is a guideline, not a hard limit — you can proceed with any configuration. But if you can address the warnings, the optimizer will converge faster.
Detail Articles
Each article below explains the reasoning behind these recommendations and gives practical examples:
Designing Your Parameter Space — how many parameters, descriptors, step sizes, and homogeneous vs mixed spaces
Choosing Objectives & Measurements — multi-objective strategies, derived quantities, and when to use constraints instead
Related
Constraints Overview — full guide to all constraint types
Multi-Objective Optimization — choosing between Pareto, Weighted Sum, and Hierarchy
Expert Context — how to write effective domain knowledge for the AI