Differentiable Python Cosmology Library
Project description
jax-cosmo
Finally a differentiable cosmology library, and it's in JAX!
Have a look at the GitHub issues to see what is needed or if you have any thoughts on the design, and don't hesitate to join the Gitter room for discussions.
TL;DR
This is what jax-cosmo
aims to do:
...
def likelihood(cosmo):
# Compute mean and covariance of angular Cls, for specific probes
mu, cov = jax_cosmo.angular_cl.gaussian_cl_covariance_and_mean(cosmo, ell, probes)
# Return likelihood value
return jax_cosmo.likelihood.gaussian_log_likelihood(data, mu, cov)
# Compute derivatives of the likelihood with respect to cosmological parameters
g = jax.grad(likelihood)(cosmo)
# Compute Fisher matrix of cosmological parameters
F = - jax.hessian(likelihood)(cosmo)
This is how you can compute gradients and hessians of any functions in jax-cosmo
,
all of this without any finite differences.
Check out a full example here:
Have a look at the design document to learn more about the structure of the code.
What is JAX?
JAX = NumPy + autodiff + GPU
JAX is a framework for automatic differentiation (like TensorFlow or PyTorch) but following the NumPy API, and using the GPU/TPU enable XLA backend.
What does that mean?
- You write plain Python/NumPy code, no need to learn a different language
- It runs on GPU, you don't need to do anything particular
- You can take derivatives of any quantity with respect to any parameters by automatic differentiation.
Checkout the JAX project page to learn more!
Install
jax-cosmo
is pure Python, so installing is a breeze:
$ pip install jax-cosmo
Philosophy
Here are some of the design guidelines:
- Implementation of equations should be human readable, and documentation should always live next to the implementation.
- Should always be trivially installable: external dependencies should be kept to a minimum, especially the ones that require compilation or with restrictive licenses.
- Keep API and implementation simple and intuitive, minimize user and developer surprise.
- “Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.” -Brian Kernighan, quote stolen from here.
Contributing
jax-cosmo
aims to be a community effort, contributions are most welcome and
can come in several forms
- Bug reports
- API design suggestions
- (Pull) requests for more features
- Examples and notebooks of cool things that can be done with the code
You can chime-in on any aspects of the design by proposing a PR to the design document. The issue page is a good place to start, but don't hesitate to come chat in the Gitter room.
Please take a look at the Contributing Document for more information.
This project follows the All Contributors guidelines aiming at recognizing and valorizing contributions at any levels.
Contributors ✨
Thanks goes to these wonderful people (emoji key):
Francois Lanusse 💻 |
Santiago Casas 🐛 💻 |
Austin Peel 💻 |
Minas Karamanis 💻 |
David Kirkby 💻 🐛 |
Alexandre Boucaud 💻 |
Denise Lanzieri 💻 |
jecampagne 🐛 |
Yin Li 💻 🐛 |
This project follows the all-contributors specification. Contributions of any kind welcome!
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
File details
Details for the file jax-cosmo-0.1.0.tar.gz
.
File metadata
- Download URL: jax-cosmo-0.1.0.tar.gz
- Upload date:
- Size: 287.4 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.10.9
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | a44003ff38381347059327b88e3fd820499150195acbf341b463db52bb461372 |
|
MD5 | fd3bc8164c69fb53cee3c8a9991672b4 |
|
BLAKE2b-256 | b1ff3d1c8d348ce2eb07cc990142cbd4dd2e72070dfd30e2dbe3660f1b392cd2 |