The following library provides a Python implementation of the Exploration Sampling Strategies presented in Lualdi et al. (2022).

To use this library, initialize a DoE class instance and select the desired sampling strategy. Run the sampling function with the desired dimensions (dd) and number of samples (nn) to generate your Design of Experiment.

Python 3.6 or higher with the following dependencies:

• numpy
• chaospy
• Custom modules: utils.util_functions, utils.objective_functions, utils.FpPLHS, utils.MqPLHS

Example of a 2D adaptive sampling process using the MIPT-alpha strategy.

# Define the dimension (dd, i.e. number of variables) and size (nn, i.e. max number of samples) of your DOE
dd = 2
nn = 16

# Set-up the adaptive sampling function
gen_doe_MIPT = DOE_class.DOE_Class('MIPT', repeat=random_seed, sizeOneStage=10).DOE

# Generate the DoE and store it
X = gen_doe_MIPT(dd, nn)

# Output:
array([[0.        , 0.22222222],
       [0.11111111, 0.55555556],
       [0.22222222, 0.88888889],
       [0.33333333, 0.11111111],
       [0.44444444, 0.44444444],
       [0.55555556, 0.77777778],
       [0.66666667, 0.        ],
       [0.77777778, 0.33333333],
       [0.88888889, 0.66666667],
       [1.        , 1.        ],
       [0.96932561, 0.04306806],
       [0.74154364, 0.97332512],
       [0.57959185, 0.19490412],
       [0.35492581, 0.6928599 ],
       [0.02270393, 0.80520333],
       [0.68483468, 0.57443643]])

• MIPT: MIPT-alpha
• FpPLHS: Fluttering perfect-Progressive Latin Hypercube Sampling (FpPLHS)
• MqPLHS: Monte Carlo quasi-Latin Hypercube Sampling (MqPLHS)

• floor_FF: Full-Factorial (FF)
• LHS_light: Latin Hypercube Sampling (LHS)
• sobol: Sobol Sequences
• halton: Halton Sequences

If you use part of the work consider citing:

Pietro Lualdi, Ralf Sturm & Tjark Siefkes. "Exploration-oriented sampling strategies for global surrogate modeling: A comparison between one-stage and adaptive methods", Journal of Computational Science, Volume 60, 2022 (https://doi.org/10.1016/j.jocs.2022.101501).