Design of Experiment¶

We use design of experiments (DOE) to generate the initial sample plan (X_init). Common methods include general full factorial design, completely randomized design and Latin hypercube sampling (LHS). The formula can be found in standard statistics books. We use LHS heavily since its near-random design and the efficient space-filling abilities.

Note

X_init is always in a unit scale.

Setting the seed parameter in random designs could make sure that the same set of data are generated every time.

Example¶

The examples are for a 3-dimensional system. Generate a full factorial design with 5 levels in each dimension.

from nextorch import doe

n_ff_level = 5
X_init_ff = doe.full_factorial([n_ff_level, n_ff_level, n_ff_level])

Generate a LHS design with 10 initial points.

from nextorch import doe

X_init_lhs = doe.latin_hypercube(n_dim=3, n_points=10, seed=1)

Generate a completely random design with 50 initial points.

from nextorch import doe

X_init_random = doe.randomized_design(n_dim=3, n_points=50, seed=1)

Here is a list of DOE functions in nextorch.doe module.

`full_factorial`	Generates full factorial design
`latin_hypercube`	Generates latin hypercube design
`randomized_design`	Generates completely randomized design
`randomized_design_w_levels`	Generates randomized design with levels in each dimension