[![Build][build-badge]][build-link] ![License: LGPL v3][license-badge]
Simcal is a simulation calibration framework for calibrating arbitrary simulators using arbitrary optimization algorithms.
Simcal provides a Simulator wrapper that provides a standardized way to invoke an arbitrary simulator. This wrapper takes a set of parameter values that instantiate the simulator's behavior, invokes the instantiated simulator one one or more input, and then returns a value that represents the simulation accuracy/quality that was achieved using these parameter values. This value is typically based on some notion of loss when compared to some ground-truth data (the details of which are left to the implementer).
Simcal also provides a Calibrator class that can be used to calibrate a simulator using some optimization algorithm. A Calibrator is expected to call a Simulator wrapper several times with different parameter values and return the parameter values that lead to the best results (lowest loss). A calibrator is provided specifications that, for each Parameter, define a type, a value range, and a distribution.
Additionally Simcal provides helpful utilities, such as an Environment for managing temporary files and command-line invocations, and a Coordinator for managing parallelism.
This package can be installed using pip install ..
Alternatively, you can set it manually up in a virtual environment using the setup.sh script.
The examples/ directory contains examples simulators and calibrators for these simulators. A complete walkthrough is provided in examples/walkthrough/walkthrough.md, and can be downloaded as an interactive jupyter notebook.