SOLikeT is a centralized package for likelihood and theory implementations for the Simons Observatory. For more extensive details please see our main documentation pages at: http://soliket.readthedocs.io/.
For a set of detailed requirements and installation instructions for different machines (e.g. NERSC, M1 Mac), please see the installation page.
SOLikeT can be installed using any of three methods: uv, pip, or conda, with the recommended method being uv.
The recommended method is uv, which offers fast, reproducible installations using a lockfile and can manage virtual environments automatically. It is extremely flexible offering complete control over the environment and dependencies. Indeed, uv substitutes for pip and conda in many cases, providing a unified interface for package management. Still, it can be used seamlessly within a conda environment and is compatible with existing pip and conda workflows.
One of the key features of uv is producing a lockfile (uv.lock) that captures the exact versions of all installed packages, ensuring reproducibility across different platforms and Python versions. This is particularly useful to avoid the "it works on my machine" problem, as it allows you to recreate the same environment on any machine with a single command.
Installing uv is as simple as running:
pip install uvThen, after activating an existing environment (both conda or venv environments are supported), you can clone the repository and install it via:
git clone https://github.com/simonsobs/soliket
cd soliket
uv sync --lockeduv sync --locked will install all the necessary dependencies to the fixed versions stored in the uv.lock file. This ensures that you have a consistent and reproducible environment, identical to the one developed and tested.
At this point, you can forget about uv, if you want to, and continue using SOLikeT as desired. If you want to re-sync your environment with the lockfile, you can re-run that command.
We define extra dependencies for SOLikeT, which allow you to install additional features or functionalities as needed. In particuar, we define emulator, pyccl, pyhalomodel, and all. Repsectively, these extras will install the CosmoPower emulator, the PyCCL library for cosmological calculations, and the PyHaloModel library for halo modeling. The last one will install all of them. These can be specified when running uv sync to install the corresponding dependencies.
uv sync --locked --extra emulator # for CosmoPower emulator
uv sync --locked --extra pyccl # for PyCCL library
uv sync --locked --extra pyhalomodel # for PyHaloModel library
uv sync --locked --extra all # for all extrasOf course, you can combine multiple extras as needed. Note that these extra dependencies come with extra constraints, so these may not be compatible with all Python versions or platforms. These are reported in the pyproject.toml file, which is used by uv to manage dependencies.
If you are less worried about reproducibility, you can also install SOLikeT without the lockfile by using pip after cloning the repository:
git clone https://github.com/simonsobs/soliket
cd soliket
pip install .In this case, you can also specify extras, such as the emulator to install the dependencies related to CosmoPower:
pip install .[emulator]For further details on uv and alternatives ways to install SOLikeT, please refer to the installation page.
SOLikeT is a collection of modules for use within the Cobaya cosmological inference and sampling workflow manager. Please see the Cobaya documentation for detailed instructions on how to use Cobaya to perform cosmological calculations and generate constraints on cosmological parameters.
SOLikeT examples and explanatory notebooks are under construction, but will be run using standard [yaml](https://en.wikipedia.org/wiki/YAML) format (which can in turn be read in as Python dictionaries). The examples will be run using something similar to:
.. code-block:: bash
cobaya-run examples/example_1.yaml
If you wish to develop your own Theory and Likelihood codes for use in SOLikeT please see the detailed instructions on the Developer Guidelines page.
Tests run a set of SOLikeT calculations with known expected results. There are (at least) two reasons you might want to run tests: verify your installation is working correclty, or check that your code changes do not break existing functionality.
To see if codes you have written when developing SOLikeT are valid and will pass the Continuous Integration (CI) tests which we require for merging on github.
If you are using uv, the easiest way to run tests (and the way we run them) is to use:
.. code-block:: bash
uv run pytest -vv --durations=10
-vv will give you verbose output, and --durations=10 will show you the 10 slowest tests, which can help identify performance issues. uv will automatically use the current environment, so you don't need to worry about activating a specific virtual environment.
If the current environment does not have the required dependencies, uv will install them automatically based on the uv.lock file, ensuring that you have all the necessary packages to run the tests.
You can also test a subset of tests or run specific tests by passing additional arguments to pytest. For example, if you want to run only the tests in a specific module, you can do
uv run pytest -vv --durations=10 -k my_new_modulesearching for tests that match the string 'my_new_module'.
If you want to run the tests using pytest directly, you can do so by running:
pytest -vv soliketThis will run the tests in the same way as uv, but without the additional features provided by uv. Note that you will need to have all the required dependencies installed in your current environment for this to work.
Indeed, running tests after installing SOLikeT in any environment is a good practice to ensure that everything is working as expected (see the installation instructions).
Please raise an issue if you have trouble installing or any of the tests fail.