Installing CaliPy

Overview

CaliPy is pre‑alpha software. You can either

  1. install the latest stable wheel from PyPI (recommended for users), or

  2. clone the development branch from GitHub (recommended for contributors and bleeding‑edge features).

Requirements

  • Python ≥ 3.8 (3.11 tested)  ⎯ 64‑bit only

  • PyTorch ≥ 2.2 with CPU or CUDA ⩾ 11.8

  • Pyro ≥ 1.9.0

  • Optional: JupyterLab or VS Code for the example notebooks.

Quick install (PyPI)

# create & activate a fresh environment (recommended)
 mkdir ~/Desktop/calipy
 cd ~/Desktop/calipy
 python3 -m venv calipy_env
 source calipy_env/bin/activate

# ... then install calipy-ppl
# (ppl = probabilistic programming language)
pip install calipy-ppl

That’s it! Verify the installation:

>>> import calipy
>>> print(calipy.__version__)

Bleeding‑edge install (GitHub)

(calipy) $ git clone https://github.com/atlasoptimization/calipy.git
(calipy) $ cd calipy
(calipy) $ pip install -e '.[dev,docs]'

The -e flag installs CaliPy in editable mode so local changes are picked up immediately—handy when you extend the library.

Optional CUDA wheels

If you have an NVIDIA GPU:

  1. Check your CUDA version with nvidia-smi.

  2. Install the matching PyTorch wheel, e.g.

    (calipy) $ pip install 'torch>=2.2+cu121' --index-url https://download.pytorch.org/whl/cu121

  3. Re‑install CaliPy (same pip install calipy[docs]).

Building the documentation locally

(calipy) $ cd calipy/docs
(calipy) $ make clean html      # outputs to build/html/index.html
(calipy) $ open build/html/index.html  # macOS; use xdg-open on Linux

Troubleshooting

  • Mismatched PyTorch/Pyro versions Make sure the Pyro release you pick supports your installed PyTorch.

  • CUDA libraries not found Install the matching CUDA toolkit for your PyTorch wheel, or fall back to the CPU wheel.

Next steps