Differentiable likelihoods through surrogate models. Our logo represents a neural network surrogate model (ghola) for a theory model (velocileptors in this case).
Neural network surrogate models and data from https://arxiv.org/abs/2510.18981 will be made available upon acceptance of this work. For early access, please contact [email protected].
h5pyfor reading emulator datanumpy,scipympi4pyfor running multiple chains simultaneouslyjax,jaxlibblackjaxfor sampling algorithmsoptaxfor minimizationinterpaxfor theory calculations
At NERSC you can run sh setup_nersc_env.sh and this should create a functional conda environment,
that you can activate as follows:
module load python
mamba activate gholaxThis is equipped with a jupyter kernel named gholax that you can use with NERSC's jupyterlab.
Analogously, assuming you have mamba installed, you can run sh setup_env.sh and it will build a functioning environment with gholax installed.
The environment can then be activated by calling mamba activate gholax.
From the repository root, with the gholax environment activated:
pytest testsThe analytic marginalization (AM) tests are slow; to skip them:
pytest tests -k "not am_"