modOpt is a modular development environment and library for optimization algorithms, written in Python. It is a platform designed to support research and education in the field of numerical optimization. Its modular development environment facilitates the construction of optimization algorithms using self-contained modules. When implementing new algorithms, developers can reuse stable and efficient modules already available in modOpt, eliminating the need to build these components from scratch. Similarly, existing algorithms in modOpt can be customized for specific applications by modifying only the relevant modules.
modOpt as a library includes several gradient-based and gradient-free optimization algorithms. It provides interfaces to more than a dozen general-purpose optimizers, along with fully transparent implementations of several educational optimization algorithms. Additionally, modOpt offers various features to support students, optimization practitioners, and advanced developers. For instance, it includes built-in visualization and recording capabilities, interfaces to modeling frameworks such as JAX, CasADi, OpenMDAO and CSDL, and an interface to the CUTEst test problem set. It also provides various utilities for testing and benchmarking algorithms, and postprocessing optimization results.
modOpt is supported on Linux, macOS, and Windows.
The general-purpose optimizers available in modOpt include SLSQP, PySLSQP,
SNOPT, IPOPT, Trust-Constr, SQP, BFGS, L-BFGS-B, Nelder-Mead, COBYLA, COBYQA, and CVXOPT.
The ConvexQPSolvers optimizer provides an interface to more than 15 QP solvers
available through the qpsolvers package.
Note that PySLSQP, SNOPT, IPOPT, COBYQA, CVXOPT, and qpsolvers must be
installed separately if users wish to utilize them.
Similarly, the modeling frameworks JAX, CasADi, OpenMDAO, and CSDL
need to be installed separately, if needed.
To install the latest commit from the main branch, run the following command in the terminal:
pip install git+https://github.com/lsdolab/modopt.git@mainTo install modOpt with all interfaced open-source optimizers, run:
pip install "modopt[open_source_optimizers] @ git+https://github.com/lsdolab/modopt.git@main"This will install pyslsqp, IPOPT, cobyqa, cvxopt, and the qpsolvers package,
along with the QP solvers quadprog and osqp.
For details on obtaining a copy of SNOPT, visit the
SNOPT documentation.
To install modOpt with JAX, CasADi, OpenMDAO, CSDL, and CSDL_alpha, run:
pip install "modopt[jax,casadi,openmdao,csdl,csdl_alpha] @ git+https://github.com/lsdolab/modopt.git@main"Remove any dependencies from the list [jax,casadi,openmdao,csdl,csdl_alpha] above if they are not needed.
To uninstall modOpt, run:
pip uninstall modoptTo upgrade to the latest commit, uninstall modOpt and then reinstall it using:
pip uninstall modopt
pip install git+https://github.com/lsdolab/modopt.git@mainTo install modOpt in development mode, clone the repository and install it using:
git clone https://github.com/lsdolab/modopt.git
pip install -e ./modoptThe -e flag installs the package in editable mode,
allowing you to modify the source code without needing to reinstall it.
To upgrade to the latest commit in development mode, navigate to the modOpt directory and run:
git pullTo verify that the installed package works correctly, install pytest using:
pip install pytestThen, run the following command from the project's root directory:
pytest -m basicThe -m basic flag runs only the basic test cases, excluding
visualization tests and tests for interfaces with
optional dependencies such as JAX, CSDL, OpenMDAO, and others.
The example below is provided to help users get started with modOpt.
For information on more advanced features, refer to the
documentation.
The following example minimizes
import numpy as np
import modopt as mo
# `v = [x, y]` is the vector of variables
# Define the problem functions
def objective(v): # Objective function
return v[0]**2 + v[1]**2
def constraint(v): # Constraint function
return np.array([v[0] + v[1]])
# Define the problem constants
x0 = np.array([1., 1.]) # Initial guess for the variables
cl = np.array([1.]) # Vector of lower bounds for the constraints
cu = np.array([1.]) # Vector of upper bounds for the constraints
# Create the problem and optimizer objects
problem = mo.ProblemLite(x0=x0, obj=objective, con=constraint, cl=cl, cu=cu)
optimizer = mo.SLSQP(problem, solver_options={'ftol':1e-6})
# Solve the optimization problem and view the results
optimizer.solve()
optimizer.print_results()We used the SLSQP optimizer to solve this problem.
Note that we did not provide functions for computing the objective gradient
or the constraint Jacobian.
In the absence of user-provided derivatives, ProblemLite estimates them
using first-order finite differences.
However, it is more efficient if the user provides the functions for the exact derivatives.
For more complex examples, such as building models in various modeling languages, using different optimizers, or developing new optimizers in modOpt with built-in modules, visit the Examples section of the documentation.
For API reference and more details on installation and usage, visit the documentation.
If you use modOpt in your work, please use the following reference for citation:
@article{joshy2024modopt,
title={modOpt: A modular development environment and library for optimization algorithms},
author={Joshy, Anugrah Jo and Hwang, John T},
journal={arXiv preprint arXiv:2410.12942},
year={2024},
doi={10.48550/arXiv.2410.12942}
}
Please use the GitHub issue tracker for reporting bugs, requesting new features, or any other questions.
We always welcome contributions to modOpt.
Please refer the CONTRIBUTING.md
file for guidelines on how to contribute.
This project is licensed under the terms of the GNU Lesser General Public License v3.0.