ert 2.36.0

Ensemble based Reservoir Tool (ERT)

ert

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).

Prerequisites

Python 3.6+ with development headers.

Installation

$ pip install ert
$ ert --help

or, for the latest development version:

$ pip install git+https://github.com/equinor/ert.git@master
$ ert --help

The ert program is based on two different repositories:

1. ecl which contains utilities to read and write Eclipse files.

2. ert - this repository - the actual application and all of the GUI.

ERT is now Python 3 only. The last Python 2 compatible release is 2.14

Documentation

Documentation for ert is located at https://ert.readthedocs.io/en/latest/.

Developing

ERT uses Python for user-facing code and C++ for some backend code. Python is the easiest to work with and is likely what most developers will work with.

Developing 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 .

Trouble with setup

If you encounter problems during install and attempt to fix them, it might be wise to delete the _skbuild folder before retrying an install.

Additional development packages must be installed to run the test suite:

pip install -r dev-requirements.txt
pytest tests/

As a simple test of your ert installation, you may try to run one of the examples, for instance:

cd test-data/local/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).

Developing C++

C++ is the backbone of ERT 2 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 (res/_lib.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. This means that, for example, Eigen computations will be checked and will abort if preconditions aren't met (eg. when inverting a matrix, it will first check that the matrix is square). 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.

Notes

1. If pip reinstallation fails during the compilation step, try removing the _skbuild directory.

2. 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 he command ulimit -a. In order to increase maximum number of open files, run ulimit -n 16384 (or some other large number) and put the command in your .profile to make it persist.

Testing C code

Install ecl using CMake as a C library. Then:

$ mkdir build
$ cd build
$ cmake ../libres -DBUILD_TESTS=ON
$ cmake --build .
$ ctest --output-on-failure

Building

Use the following commands to start developing from a clean virtualenv

$ pip install -r requirements.txt
$ python setup.py develop

Alternatively, pip install -e . will also setup ERT for development, but it will be more difficult to recompile the C library.

scikit-build is used for compiling the C library. It creates a directory named _skbuild which is reused upon future invocations of either python setup.py develop, or python setup.py build_ext. The latter only rebuilds the C library. In some cases this directory must be removed in order for compilation to succeed.

The C library files get installed into res/.libs, which is where the res module will look for them.

Example usage

Basic ert test

To test if ert itself is working, go to test-data/local/poly_example and start ert by running poly.ert with ert gui

cd test-data/local/poly_example
ert gui poly.ert

This opens up the ert graphical user interface. Finally, test ert by starting and successfully running the simulation.

ert with a reservoir simulator

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.