Installation¶

pip install emu_pk

That is the whole thing for using an emulator: numpy, jax and jaxlib, all of which have wheels on every supported platform. No conda environment is needed, and neither is a compiler.

The extras¶

	Installs	For
`emu_pk`	numpy, jax	evaluating a trained network and the correction table
`emu_pk[gen]`	+ `classy`	generating training data, and validating
`emu_pk[train]`	+ `optax`	training a network
`emu_pk[dev]`	+ pytest	running the test suite
`emu_pk[docs]`	+ sphinx	building this documentation

The split is load-bearing. import emu_pk in an environment with no classy and no optax must work, and the test suite asserts it. That is what lets another package depend on this one without inheriting a Boltzmann solver or a training stack.

A dedicated conda environment¶

The repository carries an environment.yml for a minimal environment — python, numpy and JAX, and nothing else:

mamba env create -f environment.yml     # or: conda env create -f environment.yml
mamba activate emu_pk
pip install -e .

It pins the CPU build of jaxlib, which is 64 MB against 199 MB for the CUDA one; left unpinned, the build you get depends on whether the machine that solved the environment had a driver. The file says how to swap it for a GPU, and carries a commented block per extra.

A note on `[gen]`¶

classy compiles CLASS from source, so it needs a C compiler. It is also the one part of this that can fail for environmental reasons: CLASS’s setup.py invokes make with an unbounded -j, which on a memory-capped machine gets the compiler killed —

g++: fatal error: Killed signal terminated program cc1plus

If you hit that, build on a machine without a per-user memory cap, or install classy separately with a bounded parallelism. Note that training does not need CLASS; only generation and validation do.

Supported versions¶

Python 3.11 and newer. emu_pk ships inline type information (py.typed).

Verifying the install¶

python -c "from emu_pk import PkEmulator; print(PkEmulator().pk([0.1], 0.0, \
    [0.02237, 0.12, 0.6736, 0.9649, 3.044, 0.06, -1.0, 0.0]))"

The trained weights and the correction table ship inside the package — about 18 MB — so there is no download step and the package works offline.