Installation

pip install class-e

From source, with uv:

git clone git@github.com:rcalderonb6/class-e.git
cd class-e
uv sync --all-extras          # everything, plus the development tools
uv sync --extra jax --extra classy --no-default-groups   # a lean install

Extras

Extra	Adds	Needed for
plot	cplots[matplotlib]	evaluate --plots/--report, learning and PCA curves
ui	marimo, ipywidgets, polars (includes plot)	generate -i, train -i, inspect
jax	jax, optax, tinygp, equinox	jax and vae backends, gradients
jax-cuda	jax[cuda12] on Linux (includes jax)	training on NVIDIA GPUs (GPU training)
classy	classy from PyPI	computing spectra with standard CLASS
cobaya	Cobaya	class_e.cobaya.ClassE
hf	huggingface_hub	class-e publish --to hf (downloading from Hugging Face needs nothing extra)
all	all of the above	everything

Warning

uv sync makes the environment match the command exactly: an extra left out is removed, and so is a package installed by hand with uv pip install. List every extra you want in one command, or use --all-extras.

Development groups

Tools for working on class-e itself; uv only, never published to PyPI.

Group	Adds
test	pytest
lint	ruff, codespell
docs	mkdocs, mkdocs-material, mkdocstrings, mkdocs-jupyter
notebooks	ipykernel, nbformat, nbconvert (dev/examples/*.ipynb)
dev	test, lint and docs; installed by default with uv sync

Add a group with --group notebooks; skip the default one with --no-default-groups.

CLASS

class-e computes with the classy Python wrapper. Install it in the same environment, or point class-e at another build (AxiCLASS, Hi-CLASS, ...): see Custom CLASS builds.