Optimization framework for Daisy

A python based framework for optimizing parameters in Daisy.

Features

• Multiple optimization methods: a greedy sequential optimizer, CMA-ES, and Bayesian optimizers
• Optimization across multiple scenarios
• Single or multi-objective optimization
• Optimization of parameters in both Daisy (.dai) and Python (.py) files
• Support for categorical and continuous parameters (depending on optimizer)

Getting started

The framework is available on https://pypi.org/ and can be installed with pip

pip install daisypy-optim

By default this will only install the greedy sequential optimizer. For most problems you will want to use either CMA-ES or a Bayesian optimizer. The CMA-ES optimizer can be installed with

pip install daisypy-optim[cma]

and the Bayesian optimizers with

pip install daisypy-optim[ax]

Note that the Bayesian optimizers pulls in a lot of dependencies.

Examples

There are several examples illustrating how to optimize parameters in various situations. They all share a common structure

1. Setup a DaisyRunner that knows how to run Daisy
2. Setup a FileGenerator that can generate .dai and .py
3. Define the parameters that should be optimized for
4. Define the objective that should be optimized
5. Setup and run an optimizer

See doc/examples for an overview of the examples.