epaillas / desilike

desilike

desilike is an attempt to provide a common framework for writing DESI likelihoods, that can be imported in common cosmological inference codes (Cobaya, CosmoSIS, MontePython).

desilike has the following structure:

• root directory: definition of parameters, base calculator classes, differentiation and Fisher routines, installation routines
• theories: e.g. BAO, full-shape theory models
• observables: e.g. power spectrum, correlation function
• likelihoods: e.g. Gaussian likelihood of observables, a few external likelihoods (Pantheon, Planck)
• bindings: automatic linkage with cobaya, cosmosis, montepython
• emulators: emulate e.g. full-shape theory models, to speed up inference
• samples: define chains, profiles data structures and plotting routines
• samplers: many samplers for posterior sampling
• profilers: profilers for posterior profiling

samples, samplers and profilers are provided for self-contained sampling / profiling of provided likelihoods. Example notebooks presenting most use cases are provided in directory nb/.

TODO

When autodiff is available, pass gradient to profilers / samplers whenever relavant. For ParameterArray, define when output should be a ParameterArray (e.g. under affine transforms) or not.

Documentation

Documentation in construction on Read the Docs, desilike docs. See in particular getting started.

Requirements

Only strict requirements are:

• numpy
• scipy
• pyyaml
• mpi4py
• cosmoprimo (currently with pyclass to compute DESI fiducial cosmology)

Should be made optional in the future:

• mpi4py
• pyclass (by extending TabulatedDESI to power spectra)

Installation

pip

Simply run:

python -m pip install git+https://github.com/cosmodesi/desilike

If you wish to use plotting routines (getdist, anesthetic), and tabulate for pretty tables:

python -m pip install git+https://github.com/cosmodesi/desilike#egg=desilike[plotting]

If you addtionally wish to be able to use analytic marginalization with jax:

python -m pip install git+https://github.com/cosmodesi/desilike#egg=desilike[plotting,jax]

git

First:

git clone https://github.com/cosmodesi/desilike.git

To install the code:

python setup.py install --user

Or in development mode (any change to Python code will take place immediately):

python setup.py develop --user

Other dependencies (theory codes, etc.)

Just define your calculator (most commonly your likelihood), then in a python script:

from desilike import Installer
Installer(user=True)(likelihood)

License

desilike is free software distributed under a BSD3 license. For details see the LICENSE.

Acknowledgments

• Stephen Chen, Mark Maus, Martin White for velocileptors wrapper: https://github.com/sfschen/velocileptors, https://github.com/martinjameswhite/CobayaLSS
• Pierre Zhang, Cullan Howlett, Yan Xiang Lai for pybird wrapper: https://github.com/pierrexyz/pybird, https://github.com/CullanHowlett/pybird
• Hernan E. Noriega, Alejandro Aviles for folps wrapper: https://github.com/henoriega/FOLPS-nu
• Samuel Brieden, Hector Gil-Marin, Mark Maus for ShapeFit: https://arxiv.org/abs/2106.07641
• Stephen Chen, Mark Maus for Taylor expansion emulator: https://github.com/sfschen/velocileptors_shapefit
• Stephen Chen, Joe DeRose for MLP emulator: https://github.com/sfschen/EmulateLSS
• Pat McDonald, Eva Maria Mueller, Antony Lewis for thoughts
• Pat McDonald, Edmond Chaussidon, Uendert Andrade, Daniel Forero Sanchez, Batia Friedman-Shaw, Svyatoslav Trusov, Nathan Findlay, Enrique Paillas, Vincenzo Aronica for early debugging and feedback
• Ruiyang Zhao for systematics templates
• Benedict Bahr-Kalus for turnover scale analysis: https://arxiv.org/pdf/2302.07484.pdf
• Rodrigo Calderón for Pantheon+ with/out SH0ES and Union3 likelihoods
• Cobaya, CosmoSIS bindings inspired by firecrown: https://github.com/LSSTDESC/firecrown
• Inspiration from Cobaya: https://github.com/CobayaSampler/cobaya