Warning

OpenMDAO 1.x is in an ALPHA state. The version that you are installing is under active development, and as such may be broken from time to time. Therefore, OpenMDAO 1.0 Alpha should be used at your own risk!

Intro

Purpose

This document exists to explain what OpenMDAO is, how to get it, and how to install it on OS X, Windows or Linux. For a guide of basics, tutorials and examples of how to use OpenMDAO, see the OpenMDAO User Guide.

TL;DR

Install Git, Python, Pip, Numpy, and Scipy. (Anaconda Python, comes bundled with everything you need). Next, install OpenMDAO with pip.

To get most recent release of OpenMDAO:

pip install openmdao

To get the latest commit to OpenMDAO’s Github master:

pip install git+http://github.com/OpenMDAO/OpenMDAO.git@master

This will at least get you started, but you should read the rest of this guide– we worked really hard on it!

What is OpenMDAO?

OpenMDAO is a high-performance computing platform for systems analysis and optimization that enables you to decompose your models, making them easier to build and maintain, while still solving them in a tightly-coupled manner with efficient parallel numerical methods.

We provide a library of sparse solvers and optimizers designed to work with our MPI-based, distributed-memory data-passing scheme. But don’t worry about installing MPI when you’re just getting started. We can run really efficiently in serial using a Numpy data-passing implementation as well.

Our most unique capability is our automatic analytic multidisciplinary derivatives. Provide analytic derivatives for each of your components, and OpenMDAO will solve the chain-rule across your entire model, to compute system- level derivatives for Newton solvers and/or gradient-based optimizers. This lets you solve really large non-linear problems, like a cubesat design with over 25,000 design variables using adjoint derivatives.

You don’t have to provide analytic derivatives for all of the components. OpenMDAO just finite-differences components that are missing them and then computes semi-analytic multidisciplinary derivatives. Semi-analytic derivatives offer a fast and easy way to gain a lot of computational efficiency. For example, they gave us a 5x reduction in compute cost for an aero-structural wind turbine optimization.

What’s New in OpenMDAO 1.0?

If you’re new to OpenMDAO, then all you need to know is that the API in 1.0 is different than in any older version. So, if you look at older models in a forum post or something, don’t be surprised when the code doesn’t look quite right.

If you’re an existing OpenMDAO user trying to move your models up into this version, then there are a bunch of API changes you need to be aware of. OpenMDAO 1.0 Alpha, is a departure from the versions that preceded it (OpenMDAO 0.0.1 through 0.13.0). In fact, OpenMDAO 1.0 is a complete re-write of the framework from the ground up. The new code base is much smaller (it’s now ~5000 lines of code), and will be much easier for others to work with.

We tried to follow a couple of guiding principles in the re-write:

  1. The code base should have as few dependencies as possible
  2. The user-facing API should be as small and self-consistent as possible
  3. There should not be any “magic” happening without the user’s knowledge

To that end we’ve made some significant changes to the framework. Some of the most notable differences include:

  • The Component execute method is now named solve_nonlinear
  • There are now three types of variables: parameter, output, and state
  • The way you define and access framework variables (we’ve removed our dependency on Traits)
  • The way you group sets of components together (Assembly has been replaced with Group)
  • Solvers are not drivers any more. Optimizers and DOE are still drivers
  • There is no more workflow

If you’d like to get a more direct comparison between old and new OpenMDAO input files, check out our guide to converting your models, the Pre-1.0 Conversion Guide. While a lot of the API has changed, the overall concepts are mostly the same. You still have components, which get grouped together with connections, defining data passing between them. The new API helps draw a sharper line between what is a framework variable and what is a regular Python attribute. The new API also reduces the number of different kinds of objects you have to interact with.

Since this is still an Alpha release, there is a lot of missing functionality compared to the older versions. For example, we don’t have full support for file-wrapped components yet. We’ll be working on adding the missing features as we go, but the Alpha is already very capable, especially for gradient-based optimization with analytic derivatives. We’re putting it out specifically for our users to try the new API so that they can start to play around with it. If you have any feedback, we’d love to hear it.

Installation

Quickstart

Assuming you are on a supported platform (discussed below), with the proper prerequisites installed (discussed further below), you can quickly and easily install the latest release of OpenMDAO with this line:

pip install openmdao

Or install the most recent commit to OpenMDAO from its Github repository with this:

pip install git+http://github.com/OpenMDAO/OpenMDAO.git@master

Supported Platforms

OpenMDAO will run on specified versions of Windows, Mac OS X, and Linux. However, we can’t support every version of every OS. So while you may very well be able to get OpenMDAO to run on a iPhone that’s running a Windows 3.1 emulator, we’re not going to be able to help you when something goes awry with that install. Here are the systems which we will support:

Mac OS X

  • Mountain Lion (10.8.5)
  • Mavericks (10.9.5)
  • Yosemite (10.10.4)

Linux

While we have seen successful installations using RHEL and Mint, the distribution that we use is Ubuntu. The versions of Ubuntu that we will support are:

  • Trusty Tahr (14.04.2 LTS)
  • Utopic Unicorn (14.10)
  • Vivid Vervet (15.04)

Windows

  • Windows 7
  • Windows 8

OpenMDAO Prerequisites

In order to use OpenMDAO, you will need Python installed on your system. You’ll also need a few other basic scientific computing libraries for python: Numpy and Scipy.

Note

If you want a bundled Python installation that has all our prerequisites included, try Anaconda.

Python

Currently, we are supporting two different versions of Python:

  • 2.7.9 or upper versions of 2.7.x
  • 3.4.3 or upper versions of 3.4.x

Numpy

Install Numpy, unless you already have a distribution like Anaconda that includes Numpy.

  • Version 1.9.2 or upper will be supported.

Scipy

Install Scipy, unless you already have a distribution like Anaconda that includes Scipy.

  • Version 0.15.1 or upper will be supported.

Git (Optional)

Git is a very popular open-source version control system that we use for our source code. It tracks content such as files and directories. OpenMDAO hosts its repo on GitHub. Git is not a hard requirement, but it’s a good way to stay up to date with the latest code updates (remember, we’re still in ALPHA!).

Compilers (Optional)

OpenMDAO doesn’t have a strict requirement on any compiled code, but we can optionally make use of some compiled libraries, if they are present in your Python environment. If you don’t want to use any of these optional features, then you won’t need a compiler. You can always install the compilers later and build the libraries then, and OpenMDAO will use them.

We can link to both the PyOpt and PyOpt-Sparse optimization libraries. Also, in order to run things in parallel, you’ll need petsc4py and mpi4py. So if you want to use those packages, you’ll either need platform-specific binaries for them, or you’ll need a compiler.

Install OpenMDAO Using pip

To pip install OpenMDAO’s most recent release from the Python Package Index (pypi):

pip install openmdao

To pip install OpenMDAO directly from the OpenMDAO Github repository:

pip install git+http://github.com/OpenMDAO/OpenMDAO.git@master

Clone the Repo and Install From Source (Optional)

Since the code is in ALPHA state, and is changing daily, you might prefer to actually clone our repository and install from that. This way you can always pull down the latest changes without re-installing.

git clone http://github.com/OpenMDAO/OpenMDAO.git

Then you’re going to use pip to install in development mode. Change directories to the top level of the OpenMDAO repository, and use the following command:

pip install -e .

Testing

You can run our test suite to see if your installation is working correctly. Run any single test manually by simply passing the test file to python, or you can use a test-runner, like nosetest to run the whole OpenMDAO test suite at once. Once you’ve installed nosetest, go to the top of the OpenMDAO repo and run:

nosetests .