sn-nacl 0.6.0

SN modeling package

NaCl

:warning: This code is still under development and cannot be considered stable

The NaCl package contains code to develop and train type Ia supernova spectrophotometric models. NaCl can train hybrid models, i.e. models trained simultaneously on lightcurves, spectral and/or, optionaly, spectrophotometric data, such as the spectrophotometric sequences published by the SNfactory collaboration.

As of today, NaCl contains:

• a re-implementation of the SALT2 model and error model (Guy et al, 2007, 2010), with various improvements to make sure that the training can be performed in one single minimization

• classes to manage a hybrid (light curves + spectra + photometric spectra) training sample

• a minimization framework that is able to minize a log-likelihood function, along with quadratic constraints (optional) and quadratic regularization penalities (optional). It is also possible to fit a fairly general error model simultaneously.

Table of Contents

• Installation
• [Getting started](#getting started)
• Troubleshooting
• License

Installation

virtual environments

We recommend using conda which comes with a compiled version of suitesparse. venv is also a suitable option if suitesparse is already installed on your machine, or if you are ready to compile it yourself (on Debian/Unbuntu, just have a sudo apt install libsuitesparse-dev).

As a reminder:

conda create -n MY_ENV
conda activate MY_ENV

or:

python -m venv MY_ENV
source MY_ENV/bin/activate

Prerequisites

conda packages for sn-nacl are in preparation (but not ready yet). If you are working within a conda environnment, we recommend that you install these conda packages first:

conda install ipython numpy scipy matplotlib scikit-sparse pandas h5py pyarrow

The mkl library can enhance training speed. While it is not a requirement, we recommend installing it if you are using an Intel platform:

conda install mkl sparse_dot_mkl

Until our changes are merged in sncosmo, we use a modified version of it:

git clone https://github.com/nregnault/sncosmo.git
cd sncosmo
pip install .

Finally, cmake and a C++ compiler with OpenMP support are also required to build a Python extension. On Linux, gcc supports OpenMP out of the box. On macos an installation is required:

# for linux
conda install cmake
# for macos
conda install cmake openmp

Installing NaCl

pip install sn-nacl

or, if you prefer installing it from sources:

git clone https://gitlab.in2p3.fr/lemaitre/sn-nacl
cd sn-nacl
pip install -e .

If your you are a developper and want to work on the sn-nacl package:

pip install nanobind ninja scikit-build-core[pyproject]
pip install --no-build-isolation -Ceditable.rebuild=true -ve .

Running the tests

pip install .[test]
pytest

Building the documentation

pip install -e .[doc]
cd docs
make html  # for example, type "make" for a list of targets

Then the documenation is available at docs/build/html/index.html

Getting started

TBD

Troubleshooting

Dealing with scikit-sparse

NaCl depends on the python bindings to the cholmod package, distributed with the scikit-sparse package. In most cases, scikit-sparse will be fetched automatically and installed without any problem. However, depending on your version of python the installation may crash for various reasons:

• The package needs numpy to be installed already (and doesn't fetch it automatically if missing).

• The packages needs cython : pip install cython should solve the problem.

• Finally, libsuitesparse should also be available on your system. On a debian or ubuntu system: sudo apt-get install libsuitesparse-dev should suffice. On Fedora/CentOS the equivalent is sudo yum install libsuitesparse-devel. Otherwise, an alternative is to to the SuiteSparse repository, clone it and follow the installation instructions.

• Also, for python>=3.10, the pip installation of scikit-sparse may complain about various things and stops. If you encounter this kind of problem, this longer install sequence should work:

  pip install numpy
  git clone https://github.com/scikit-sparse/scikit-sparse.git
  cd scikit-sparse; python setup.py install; cd ..
  pip install sn-nacl
  

License

sn-nacl is distributed under the terms of the MIT license.