- You can see the full rendered docs at: https://retworkx.readthedocs.io/en/latest/index.html
retworkx is a general purpose graph library for python3 written in Rust to take advantage of the performance and safety that Rust provides. It was built as a replacement for qiskit's previous (and current) networkx usage (hence the name) but is designed to provide a high performance general purpose graph library for any python application. The project was originally started to build a faster directed graph to use as the underlying data structure for the DAG at the center of qiskit-terra's transpiler, but it has since grown to cover all the graph usage in Qiskit and other applications.
retworkx is published on pypi so on x86_64, i686, ppc64le, s390x, and aarch64 Linux systems, x86_64 on Mac OSX, and 32 and 64 bit Windows installing is as simple as running:
pip install retworkxThis will install a precompiled version of retworkx into your python environment.
If there are no precompiled binaries published for your system you'll have to
build the package from source. However, to be able able to build the package
from the published source package you need to have rust >=1.39 installed (and
also cargo which is normally included with
rust) You can use rustup (a cross platform installer for
rust) to make this simpler, or rely on
other installation methods.
A source package is also published on pypi, so you still can also run the above
pip command to install it. Once you have rust properly installed, running:
pip install retworkxwill build retworkx for your local system from the source package and install it just as it would if there was a prebuilt binary available.
The first step for building retworkx from source is to clone it locally with:
git clone https://github.com/Qiskit/retworkx.gitretworkx uses PyO3 and
setuptools-rust to build the
python interface, which enables using standard python tooling to work. So,
assuming you have rust installed, you can easily install retworkx into your
python environment using pip. Once you have a local clone of the repo, change
your current working directory to the root of the repo. Then, you can install
retworkx into your python env with:
pip install .Assuming your current working directory is still the root of the repo. Otherwise you can run:
pip install $PATH_TO_REPO_ROOTwhich will install it the same way. Then retworkx is installed in your
local python environment. There are 2 things to note when doing this
though, first if you try to run python from the repo root using this
method it will not work as you expect. There is a name conflict in the
repo root because of the local python package shim used in building the
package. Simply run your python scripts or programs using retworkx
outside of the repo root. The second issue is that any local changes you
make to the rust code will not be reflected live in your python environment,
you'll need to recompile retworkx by rerunning pip install to have any
changes reflected in your python environment.
If you'd like to build retworkx in debug mode and use an interactive debugger
while working on a change you can use python setup.py develop to build
and install retworkx in develop mode. This will build retworkx without
optimizations and include debuginfo which can be handy for debugging. Do note
that installing retworkx this way will be significantly slower then using
pip install and should only be used for debugging/development.
It's worth noting that pip install -e does not work, as it will link the python
packaging shim to your python environment but not build the retworkx binary. If
you want to build retworkx in debug mode you have to use
python setup.py develop.
Once you have retworkx installed you can use it by importing retworkx. All the functions and graph classes are off the root of the package. For example, building a DAG and adding 2 nodes with an edge between them would be:
import retworkx
my_dag = retworkx.PyDAG(cycle_check=True)
# add_node(), add_child(), and add_parent() return the node index
# The sole argument here can be any python object
root_node = my_dag.add_node("MyRoot")
# The second and third arguments can be any python object
my_dag.add_child(root_node, "AChild", ["EdgeData"])