distribution-surrogates

Set up of surrogate models and order reduction for stochastic distributions

Outline

Problem and solution development

1. 1_emulator_basics: initial overview of need for distributions when using agent-based simulation, overview of emulation approach using joint distributions defined using parametric marginals and copulas, and an initial joint distribution example
2. 2_toy_data_OR: initial exploration of statistical tools and unsupervised order reduction modeling tools using a toy dataset from cancer research available through scikit-learn (detailed usage of the techniques used here will be in the following notebooks)
3. 3_emulator_nonparametric_OR: initial setup for emulator - a generative probabilistic model - considering simulation, structuring data, and testing statistical consistency
4. 4_emulator_parametric_OR: setup for parametric emulator - using latin hypercube design of experiments - instantiation of order reduction approach with parametric inputs, identification of issues with histogram-based data representation, successful use of ecdf-based representation
5. 5_emulator_predictive_modeling: connection to framework for surrogate modeling
6. 6_emulator_interactive_model: instantiation of interactive model and simulation in tandem to showcase exploration capability with surrogate model

Functions and tools

• Custom tools:
  • stats_functions
  • copula_gen_data
  • emulator_classes
• Tools from [1], [2] for surrogate modeling
  • pyToolbox [1]
  • Requires Visual Studio Compiler, specifically, build tools for visual studio 2022, selecting at least the C++ library
  • fit_(...).py, hierarchical_kriging, multifidelity_rbf, swiss_roll_test, base_classes [2]

References:

[1] [2]