Design with explicit Formula

This tutorial notebook shows how to setup a D-optimal design with BoFire while providing an explicit formula and not just one of the four available keywords linear, linear-and-interaction, linear-and-quadratic, fully-quadratic.

Make sure that cyipoptis installed. The recommend way is the installation via conda conda install -c conda-forge cyipopt.

Imports

import bofire.strategies.api as strategies
from bofire.data_models.api import Domain, Inputs
from bofire.data_models.features.api import ContinuousInput
from bofire.data_models.strategies.api import DoEStrategy
from bofire.data_models.strategies.doe import DOptimalityCriterion
from bofire.utils.doe import get_confounding_matrix

Setup of the problem

input_features = Inputs(
    features=[
        ContinuousInput(key="a", bounds=(0, 5)),
        ContinuousInput(key="b", bounds=(40, 800)),
        ContinuousInput(key="c", bounds=(80, 180)),
        ContinuousInput(key="d", bounds=(200, 800)),
    ],
)
domain = Domain(inputs=input_features)

Definition of the formula for which the optimal points should be found

model_type = "a + {a**2} + b + c + d + a:b + a:c + a:d + b:c + b:d + c:d"
model_type
'a + {a**2} + b + c + d + a:b + a:c + a:d + b:c + b:d + c:d'

Find D-optimal Design

data_model = DoEStrategy(
    domain=domain,
    criterion=DOptimalityCriterion(formula=model_type),
    ipopt_options={"max_iter": 100, "print_level": 0},
)
strategy = strategies.map(data_model=data_model)
design = strategy.ask(17)
design
/opt/hostedtoolcache/Python/3.12.13/x64/lib/python3.12/site-packages/bofire/strategies/doe/utils.py:819: OptimizeWarning: Unknown solver options: disp
  result = opt.minimize(
a b c d
0 5.000000 689.187566 180.0 800.000000
1 5.000000 800.000000 180.0 200.000000
2 0.000000 462.614227 180.0 200.000000
3 5.000000 40.000000 80.0 200.000000
4 5.000000 722.050985 80.0 200.000000
5 4.541290 40.000000 80.0 800.000000
6 5.000000 275.934846 180.0 800.000000
7 0.000000 800.000000 180.0 800.000000
8 0.000000 40.000000 80.0 539.208580
9 0.000000 40.000000 180.0 800.000000
10 0.000000 800.000000 80.0 200.000000
11 2.633123 800.000000 180.0 719.539638
12 5.000000 660.747921 180.0 200.000000
13 0.000000 800.000000 80.0 800.000000
14 5.000000 800.000000 80.0 800.000000
15 2.864761 40.000000 180.0 354.712428
16 0.000000 800.000000 180.0 638.070935

Analyze Confounding

import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns


matplotlib.rcParams["figure.dpi"] = 120

m = get_confounding_matrix(
    domain.inputs,
    design=design,
    interactions=[2, 3],
    powers=[2],
)

sns.heatmap(m, annot=True, annot_kws={"fontsize": 7}, fmt="2.1f")
plt.show()