ert - Ensemble based Reservoir Tool - is designed for running ensembles of dynamical models such as reservoir models, in order to do sensitivity analysis and data assimilation. ert supports data assimilation using the Ensemble Smoother (ES), Ensemble Smoother with Multiple Data Assimilation (ES-MDA) and Iterative Ensemble Smoother (IES).
$ pip install ert
$ ert --help
or, for the latest development version (requires Python development headers):
$ pip install git+https://github.com/equinor/ert.git@main
$ ert --help
For examples and help with configuration, see the ert Documentation.
ert was originally written in C/C++ but most new code is Python.
You might first want to make sure that some system level packages are installed before attempting setup:
- pip
- python include headers
- (python) venv
- (python) setuptools
- (python) wheel
It is left as an exercise to the reader to figure out how to install these on their respective system.
To start developing the Python code, we suggest installing ert in editable mode into a virtual environment to isolate the install (substitute the appropriate way of sourcing venv for your shell):
# Create and enable a virtualenv
python3 -m venv my_virtualenv
source my_virtualenv/bin/activate
# Update build dependencies
pip install --upgrade pip wheel setuptools
# Download and install ert
git clone https://github.com/equinor/ert
cd ert
pip install --editable .
Additional development packages must be installed to run the test suite:
pip install ".[dev]"
pytest tests/
Git LFS must be installed to get all the files. This is packaged as git-lfs
on Ubuntu, Fedora or macOS Homebrew. For Equinor RGS node users, it is possible to use git
from Red Hat Software Collections:
source /opt/rh/rh-git227/enable
test-data/block_storage is a submodule and must be checked out.
git submodule update --init --recursive
If you checked out submodules without having git lfs installed, you can force git lfs to run in all submodules with:
git submodule foreach "git lfs pull"
There are a set of style requirements, which are gathered in the pre-commit
configuration, to have it automatically run on each commit do:
$ pip install pre-commit
$ pre-commit install
If you encounter problems during install, try deleting the _skbuild
folder before reinstalling.
As a simple test of your ert
installation, you may try to run one of the
examples, for instance:
cd test-data/poly_example
# for non-gui trial run
ert test_run poly.ert
# for gui trial run
ert gui poly.ert
Note that in order to parse floating point numbers from text files correctly,
your locale must be set such that .
is the decimal separator, e.g. by setting
# export LC_NUMERIC=en_US.UTF-8
in bash (or an equivalent way of setting that environment variable for your shell).
C++ is the backbone of ert as in used extensively in important parts of ert. There's a combination of legacy code and newer refactored code. The end goal is likely that some core performance-critical functionality will be implemented in C++ and the rest of the business logic will be implemented in Python.
While running --editable
will create the necessary Python extension module
(src/ert/_clib.cpython-*.so
), changing C++ code will not take effect even when
reloading ert. This requires recompilation, which means reinstalling ert from
scratch.
To avoid recompiling already-compiled source files, we provide the
script/build
script. From a fresh virtualenv:
git clone https://github.com/equinor/ert
cd ert
script/build
This command will update pip
if necessary, install the build dependencies,
compile ert and install in editable mode, and finally install the runtime
requirements. Further invocations will only build the necessary source files. To
do a full rebuild, delete the _skbuild
directory.
Note: This will create a debug build, which is faster to compile and comes with
debugging functionality enabled. The downside is that this makes the code
unoptimised and slow. Debugging flags are therefore not present in builds of
ert that we release on Komodo or PyPI. To build a release build for development,
use script/build --release
.
-
If pip reinstallation fails during the compilation step, try removing the
_skbuild
directory. -
The default maximum number of open files is normally relatively low on MacOS and some Linux distributions. This is likely to make tests crash with mysterious error-messages. You can inspect the current limits in your shell by issuing the command
ulimit -a
. In order to increase maximum number of open files, runulimit -n 16384
(or some other large number) and put the command in your.profile
to make it persist.
The C++ code and tests require resdata. As long
as you have pip install resdata
'd into your Python virtualenv all should work.
# Create and enable a virtualenv
python3 -m venv my_virtualenv
source my_virtualenv/bin/activate
# Install build dependencies
pip install pybind11 conan cmake resdata
# Build ert and tests
mkdir build && cd build
cmake ../src/clib -DCMAKE_BUILD_TYPE=Debug
make -j$(nproc)
# Run tests
ctest --output-on-failure
To test if ert itself is working, go to test-data/poly_example
and start ert by running poly.ert
with ert gui
cd test-data/poly_example
ert gui poly.ert
This opens up the ert graphical user interface. Finally, test ert by starting and successfully running the simulation.
To actually get ert to work at your site you need to configure details about
your system; at the very least this means you must configure where your
reservoir simulator is installed. In addition you might want to configure e.g.
queue system in the site-config
file, but that is not strictly necessary for
a basic test.