Installation

ropt is distributed on PyPI and can be installed with any standard Python package manager. It requires Python 3.12 or newer.

Install the core package

Using pip:

pip install ropt

The core install includes the built-in SciPy-based optimizers and samplers, which are enough for most basic optimization tasks.

Optional extras

ropt offers a few optional dependency groups that add extra functionality:

Extra	Pulls in	Enables
`pandas`	`pandas`	Exporting results to pandas data frames.
`polars`	`polars`	Exporting results to polars data frames.
`cloudpickle`	`cloudpickle`	Copying lambdas, closures, and notebook-defined code into separate processes.
`hpc`	`pysqa`	Running evaluations on HPC clusters.

Why cloudpickle?

Some ways of running evaluations do not call your function inside your own program: they start separate processes — on your own machine, or as jobs on an HPC cluster — and run it there. Your function and its data must be copied into those processes, which Python does with its standard pickle module. pickle copies a function by storing its name, so the other process can only rebuild functions and classes that are defined at the top level of a module it can import. With cloudpickle installed, the code itself is copied, which adds lambdas, functions defined inside another function (closures), and functions written in a Jupyter notebook.

cloudpickle is optional in every case: it never changes what ropt can do, only where the copied code may be defined. Both evaluating in parallel and the external backend state what they accept without it.

Install with:

pip install "ropt[pandas]"
pip install "ropt[polars]"
pip install "ropt[pandas,hpc,cloudpickle]"

Plugin packages

Additional optimization backends are provided as standalone packages. Once installed alongside ropt, they are picked up automatically:

Package	Adds
`ropt-dakota`	Algorithms from the Dakota toolkit.
`ropt-nomad`	The MADS algorithm via NOMAD.
`ropt-pymoo`	Algorithms from `pymoo`.

Install any of them alongside ropt:

pip install ropt ropt-pymoo

After installation, the plugin's methods become available through the backend.method field in your configuration, either as "plugin/method" (for example "pymoo/nelder-mead") or, if the method name is unique among your installed plugins, just "method" (for example "nelder-mead"). See the method strings section of the configuration guide for the full details.

Verifying the installation

A quick sanity check:

# Print the current version:
from ropt.version import __version__
print(__version__)

# Verify the SciPy backend is available:
from ropt.plugins import get_plugin_name
print(get_plugin_name("backend", "slsqp"))  # should print "scipy"

If scipy is printed, the default backend plugin is working. Any additional plugin packages you installed can be verified by checking their methods in the same way.