Slycat Ensemble Analysis of Electrical Circuit Simulations

An ensemble is a group of related simulation runs, each consisting of the same set of variables, in a shared high-dimensional space describing a particular problem domain. Ensemble analysis looks at the combined behaviors and features of the simulations to discover higher-level patterns that describe aspects of the underlying problem space. Sensitivity analysis is a type of ensemble analysis that evaluates how changes in simulation input parameters correlate with simulation results. Commonly, simple regression and multiple regression techniques are used to correlate single inputs to single outputs, or groups of inputs to a single output, respectively. However, neither of these approaches evaluates the collective relationships among multiple inputs and outputs. Existing visualization tools are fundamentally designed to view no more than a few simulations in combination. Ensembles containing hundreds or thousands of simulations require a different type of analysis, different visual abstractions, and a different system architecture to effectively manage integrating so many results. We present Slycat, a scalable, collaborative, remote analysis and visualization system designed for ensemble analysis. Slycat uses Canonical Correlation Analysis (CCA) to model relationships between input and output variables, providing a generalized correlation capability that analyzes any variable subsets from the two variable groups. Using linked views, we provide multiple representations of the CCA results for an ensemble, each showing the results at a different level of abstraction. The tight integration of analysis and visualization allows analysts to iteratively explore their data, forming and testing hypotheses about how simulation input parameters are driving output results in their ensembles. We provide two real-life examples using electrical circuit simulation ensembles of differing scales to demonstrate Slycat’s utility in answering common analysis questions. Slycat is available under an open source license through github.