Model Setup#

This page describes the system requirements and installation procedures to set up ActivitySim.

Note

ActivitySim is under active development

Quick Reference#

This section briefly describes the quickest way to install and start running ActivitySim. This section assumes the user is more experienced in running travel demand models and proficient in Python, but has not used ActivitySim or has not used recent versions of ActivitySim. More detailed instructions for installing and running ActivitySim are also available in this Users Guide.

System Requirements#

This section highlights the software requirements for any implementation, as well as hardware recommendations.

Hardware#

The computing hardware required to run a model implemented in the ActivitySim framework generally depends on:

  • The number of households to be simulated for disaggregate model steps

    • In addition to the total number of households in the model region, runtime and hardware requirements can be reduced by sampling a subset of the households. The user can adjust the sampling rate for a particular run (see Settings.yaml).

  • The number of model zones (for each zone system) for aggregate model steps

  • The number and size of network skims by mode and time-of-day

  • The number of zone systems, see Zone System

  • The desired runtimes

ActivitySim framework models use a significant amount of RAM since they store data in-memory to reduce data access time in order to minimize runtime.

For example, the SEMCOG ABM, a model that follows a 2-Zone system runs on a windows machine, with the minimum and recommended system specification as follows:

  • Minimum Specification:

    • Operating System: 64-bit Windows 7, 64-bit Windows 8 (8.1) or 64-bit Windows 10

    • Processor: 8-core CPU processor

    • Memory: 128 GB RAM

    • Disk space: 150 GB

  • Recommended Specification:

    • Operating System: 64-bit Windows 7, 64-bit Windows 8 (8.1), 64-bit Windows 10, or 64-bit Windows 11

    • Processor: Intel CPU Xeon Gold / AMD CPU Threadripper Pro (12+ cores)

    • Memory: 256 GB RAM

    • Disk space: 150 GB

As another example, the prototype MTC example model - which has 2.7 million households, 7.5 million people, 1475 zones, 826 network skims - has a runtime between one hour and one day depending on the amount of RAM and number of processors allocated.

ActivitySim has features that makes it possible to customize model runs or improve model runtimes based on the available hardware resources and requirements. A few ways to do this are listed below:

  • Chunking allows the user to run eligible steps in parallel. This can be turned on/off.

  • Multiprocessing allows the user to segment processing into discrete sets of data. This will increase the runtime but allow for lower RAM requirements. This feature can also be turned on/off.

  • Sharrow is a Python library designed to decrease run-time for ActivitySim models by creating an optimized compiled version of the model. This can also be turned on/off.

  • Tracing allows the user to access information throughout the model run for a specified number of households/persons/zones. Enabling this feature will increase run-time and memory usage. It is recommended that this feature be turned off for typical model application.

  • Optimization of data types including:

    • Converting string variables to pandas categoricals. ActivitySim releases <placeholder for version number> and higher have this capability.

    • Converting higher byte integer variables to lower byte integer variables (such as reducing ‘num tours’ from int64 to int8).

    • Converting higher byte float variables to lower bytes. ActivitySim releases X.X.X and higher have this capability as a switch and defaults to turning this feature off.

Steps for enabling/disabling these options are included in the Advanced Configuration sub-section, under Ways to Run the Model page of this Users’ Guide.

Note

In general, more CPU cores and RAM will result in faster run times. ActivitySim has also been run in the cloud, on both Windows and Linux using Microsoft Azure. Example configurations, scripts, and runtimes are in the <todo: cross-ref> other_resources\example_azure folder.

Software#

Activitysim is implemented in the Python programming language. It also uses several open source Python packages such as pandas, numpy, pytables, openmatrix etc. Hence it is recommended that you install and use a conda package manager for your system. One easy way to do so is by using Mambaforge. Mamba is a free open source cross-platform package manager that runs on Windows, OS X and Linux and is fully compatible with conda packages. It is also usually substantially faster than conda itself. Instructions to install mambaforge can be found here. Installers for different Operating Systems can be found here.

Alternatively, if you prefer a package installer backed by corporate tech support available (for a fee) as necessary, you can install Anaconda 64bit Python 3, although you should consult the terms of service for this product and ensure you qualify since businesses and governments with over 200 employees do not qualify for free usage. If you’re using conda instead of mamba, just replace every call to mamba below with conda, as they share the same user interface and most command formats.

If you access the internet from behind a firewall, then you may need to configure your proxy server. To do so, create a .condarc file in your home installation folder, such as:

proxy_servers:
  http: http://myproxy.org:8080
  https: https://myproxy.org:8080
ssl_verify: false

Installing ActivitySim#

There are multiple ways to install the ActivitySim codebase:

  1. Using a Pre-packaged Installer (recommended for users who do not need to change the Python code)

  2. Using a Python package manager like mamba (recommended for users who need to change/customize the Python code)

  3. Using pip - Python’s standard package manager

Pre-packaged Installer#

Begining with version 1.2, ActivitySim is now available for Windows via a pre-packaged installer. This installer provides everything you need to run ActivitySim, including Python, all the necessary supporting packages, and ActivitySim itself. You should only choose this installation process if you plan to use ActivitySim but you don’t need or want to do other Python development. Note this installer is provided as an “executable” which (of course) installs a variety of things on your system, and it is quite likely to be flagged by Windows, anti-virus, or institutional IT policies as “unusual” software, which may require special treatment to actually install and use.

Download the installer from GitHub here. It is strongly recommended to choose the option to install “for me only”, as this should not require administrator privileges on your machine. Pay attention to the complete path of the installation location. You will need to know that path to run ActivitySim in the future, as the installer does not modify your “PATH” and the location of the ActivitySim.exe command line tool will not be available without knowing the path to where the install has happened.

Once the install is complete, ActivitySim can be run directly from any command prompt by running <install_location>/Scripts/ActivitySim.exe.

Using mamba package manager#

This method is recommended for ActivitySim users who also wish to customize the Python code to run their models. The steps involved are described as follows:

  1. Install the mamba package manager as described in the Software Requirements subsection.

2. Create a conda environment (basically a Python install just for this project) using mambaforge prompt or conda prompt depending on the package manager you use (on Windows) or the terminal (macOS or Linux):

mamba create -n asim python=3.9 activitysim -c conda-forge --override-channels

This command will create the environment and install all the dependencies required for running ActivitySim. It is only necessary to create the environment once per machine, you do not need to (re)create the environment for each session. If you would also like to install other tools or optional dependencies, it is possible to do so by adding additional libraries to this command. For example:

mamba create -n asim python=3.9 activitysim jupyterlab larch -c conda-forge --override-channels

This example installs a specific version of Python, version 3.9. A similar approach can be used to install specific versions of other libraries as well, including ActivitySim, itself. For example:

mamba create -n asim python=3.9 activitysim=1.0.2 -c conda-forge --override-channels

Additional libraries can also be installed later. You may want to consider these tools for certain development tasks:

# packages for testing
mamba install pytest pytest-cov coveralls black flake8 pytest-regressions -c conda-forge --override-channels -n asim

# packages for building documentation
mamba install sphinx numpydoc sphinx_rtd_theme==0.5.2 -c conda-forge --override-channels -n asim

# packages for estimation integration
mamba install larch -c conda-forge --override-channels -n asim

# packages for example notebooks
mamba install jupyterlab matplotlib geopandas descartes -c conda-forge --override-channels -n asim

To create an environment containing all these optional dependencies at once, you can run the shortcut command

mamba env create activitysim/ASIM -n asim

  1. To use the asim environment, you need to activate it

::

conda activate asim

The activation of the correct environment needs to be done every time you start a new session (e.g. opening a new conda Prompt window).

Note

The activate and deactivate commands to start and stop using environments are called as conda even if you are otherwise using mamba. mamba is a drop-in replacement and uses the same commands and configuration options as conda. You can swap almost all commands between conda & mamba. For more details, refer to the mamba user guide.

Using pip - Python’s standard package manager#

If you prefer to install ActivitySim without a package manager like mamba or conda, it is possible to do so with pip, although you may find it more difficult to get all of the required dependencies installed correctly. If you can use conda for the dependencies, you can get most of the libraries you need from there:

# required packages for running ActivitySim
mamba install cytoolz numpy pandas psutil pyarrow numba pytables pyyaml openmatrix requests -c conda-forge

# required for ActivitySim version 1.0.1 and earlier
pip install zbox

And then simply install activitysim with pip.

python -m pip install activitysim

If you are using a firewall you may need to add --trusted-host pypi.python.org --proxy=myproxy.org:8080 to this command.

For development work, can also install ActivitySim directly from source. Clone the ActivitySim repository, and then from within that directory run:

python -m pip install . -e

The “-e” will install in editable mode, so any changes you make to the ActivitySim code will also be reflected in your installation.

Installing from source is easier if you have all the necessary dependencies already installed in a development conda environment. Developers can create an environment that has all the optional dependencies preinstalled by running:

mamba env create activitysim/ASIM-DEV

If you prefer to use a different environment name than ASIM-DEV, just append –name OTHERNAME to the command. Then all that’s left to do is install ActivitySim itself in editable mode as described above.

Note

ActivitySim is a 64bit Python 3 library that uses a number of packages from the scientific Python ecosystem, most notably pandas and numpy.

As mentioned above, the recommended way to get your own scientific Python installation is to install 64 bit Anaconda, which contains many of the libraries upon which ActivitySim depends + some handy Python installation management tools.

Anaconda includes the conda command line tool, which does a number of useful things, including creating environments (i.e. stand-alone Python installations/instances/sandboxes) that are the recommended way to work with multiple versions of Python on one machine. Using conda environments keeps multiple Python setups from conflicting with one another.

You need to activate the activitysim environment each time you start a new command session. You can remove an environment with conda remove -n asim --all and check the current active environment with conda info -e.

For more information on Anaconda, see Anaconda’s getting started guide.