The Exploratory Modeling Workbench: An open source toolkit for exploratory modeling, scenario discovery, and (multi-objective) robust decision making

There is a growing interest in model-based decision support under deep uncertainty, reflected in a variety of approaches being put forward in the literature. A key idea shared among these is the use of models for exploratory rather than predictive purposes. Exploratory modeling aims at exploring the implications for decision making of the various presently irresolvable uncertainties. This is achieved by conducting series of computational experiments that cover how the various uncertainties might resolve. This paper presents an open source library supporting this. The Exploratory Modeling Workbench is implemented in Python. It is designed to (i) support the generation and execution of series of computational experiments; and (ii) support the visualization and analysis of the results from the computational experiments. The Exploratory Modeling Workbench enables users to easily perform exploratory modeling with existing models, identify the policy-relevant uncertainties, assess the efficacy of policy options, and iteratively improve candidate strategies. © 2017 The Author. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Software availability Name of Software: Exploratory Modeling Workbench Description: The Exploratory Modeling Workbench is an open source Python library for exploratory modeling. Exploratory modeling underpins the various modelbased approaches for decision making under deep uncertainty. The library can be used to develop interfaces to existing simulation models, define computational experiments to conduct with those models, analyze the results of these experiments, and store the results. The software is available through pip, a Python package manager: pip install ema_workbench. It depends on the Python scientific computing stack (numpy, scipy, matplotlib) as well as seaborn and ipyparallel (both are available through the conda package manager). Optional dependencies are platypus (available through github) for many-objective optimization, SALib (available through the pip package manager) for global sensitivity analysis, and mpld3 (available through conda) for interactive visualizations of PRIM td. This is an open access article u Developer: Jan H. Kwakkel ([email protected]) with contributions from M. Jaxa-Rozen, S. Eker, W. Auping, E. Pruyt, and C. Hamarat Source language: Python Supported systems: unix, linux, windows, Mac License: BSD 3 clause