When standard experiment templates like grids or boxes don't fit your constraints, use a D-optimal design. This method allows you to define a specific set of physically possible test combinations and uses an algorithm to pick the most informative runs, ensuring you get the best results from the fewest experiments.
Flexible Experimental Planning
Traditional designs often rely on rigid, pre-defined geometric patterns that may force you to run experiments which are either impossible or dangerous. D-optimal design moves away from these templates, focusing instead on the information quality of the data points you actually choose to test.
By tailoring the design to your specific candidate pool, you avoid wasting resources on unnecessary test conditions. This approach is suited for complex scenarios where experimental factors interact in non-standard ways.
Maximizing Information
The core of this method is the D-criterion, which calculates a score based on the mathematical properties of your chosen experiment matrix. A higher score represents a design that minimizes the uncertainty of your model parameters.
Lattice uses a coordinate exchange algorithm to iterate through different combinations of your candidate set. By running multiple cycles, it searches for the configuration that provides the most reliable statistical insight, ensuring your final plan is mathematically sound.
Handling Constraints and Complexity
Many engineering and scientific processes have 'forbidden zones'—combinations of factors that cause equipment damage or process failure. D-optimal design handles these by allowing you to pre-filter your candidate set, effectively removing invalid configurations before the optimization algorithm even starts.
Because this design is not tied to a specific shape, it naturally accommodates irregular levels and mixed-factor designs. Whether you are running a simple screening or a complex quadratic model, the system adapts to your unique set of requirements.
1 · Intent → method
An LLM picks doe_generate_doptimal from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
Why do I need to provide a candidate set?
A D-optimal design works by evaluating a pool of potential test combinations. By providing a candidate set, you tell the algorithm which experiments are physically possible in your specific environment, allowing it to choose the best subset that maximizes information while respecting your constraints.
How does the algorithm pick the number of runs?
The algorithm determines the minimum number of runs based on the complexity of your model (linear, interaction, or quadratic). You specify the total number of experiments you can afford to run, and the algorithm selects the most efficient points from your candidate set to satisfy that target.
Tool input schema
Schema for doe_generate_doptimal not exported yet (run pnpm export:registry).