Installation

Curie is available through the Python Package Index, which allows installation using Python’s standard command line utility pip. Assuming Python and pip are installed already, you can install Curie with the command:

pip install --user curie

or:

python -m pip install --user curie

If you’d like to install the most recent (unreleased) version of Curie, you can clone the Curie GitHub repository and install it from the source tree:

git clone https://github.com/jtmorrell/curie.git
python -m pip install --user ./curie

Each release’s built wheel is also attached to the corresponding GitHub release, for machines that cannot reach PyPI — see Installing behind a restrictive proxy or firewall below.

Nuclear data

Curie downloads the nuclear data files it needs the first time they are used: each database is fetched from the curie data release on first access, verified against its published SHA256 checksum, and stored in a per-user data directory. The large cross-section libraries (ENDF and the TENDL variants) are fetched in small per-target pieces, so looking up one reaction downloads well under a MB rather than a 37-748 MB library. No action is needed to make this happen — the first ci.Reaction(...) or ci.Isotope(...) just works.

To download everything up front (for offline use, or to pre-populate the data directory in one step):

import curie as ci
ci.download()

If your data files are stale or corrupted, re-download them with:

ci.download(overwrite=True)

The data directory follows your platform’s convention:

Windows: %LOCALAPPDATA%\curie — almost always C:\Users\<username>\AppData\Local\curie
Linux: ~/.local/share/curie
macOS: ~/Library/Application Support/curie

Data installed by Curie versions before 0.1.0 (inside the package’s own directory) is found and adopted into the data directory automatically — nothing is re-downloaded after an upgrade.

Troubleshooting

Installing behind a restrictive proxy or firewall

On some networks a proxy blocks pip, conda, and Curie’s automatic data downloads, while ordinary browser downloads from github.com still work. Curie can be installed entirely by hand in that situation — no environment variables or administrator rights are needed.

Installing the code. Download the wheel (the .whl file) from the latest GitHub release in a browser and install it locally:

pip install path/to/curie-X.Y.Z-py3-none-any.whl

Installing from a local wheel requires no network access. If pip cannot reach PyPI to resolve dependencies, add --no-deps — Curie’s dependencies (numpy, scipy, pandas, matplotlib, pooch, platformdirs) are all included in a standard Anaconda installation. Prefer the wheel over copying the source tree into site-packages by hand: a hand-copied tree has no package metadata, so pip cannot see, upgrade, or uninstall it.

Installing the data. Run any Curie command that needs data once, e.g.:

import curie as ci
ci.download()

The download will fail, but this creates the data directory, and the error message reports its location along with the data release page to fetch from. Download the .db files from the curie data release page in a browser and place them in that directory. The downloads can be verified against the SHA256SUMS file published on the same release page.

On Windows the data directory is %LOCALAPPDATA%\curie — almost always C:\Users\<username>\AppData\Local\curie. The AppData folder is hidden by default in File Explorer, but typing %LOCALAPPDATA%\curie directly into the Explorer address bar (or the Win+R run dialog) opens it.

Alternatives. On any machine that can reach GitHub, ci.download() populates the data directory; copying that directory to the same location on the restricted machine works just as well as downloading by hand. To place the data somewhere else — a shared network drive, a project directory — set the CURIE_DATA_DIR environment variable to that folder before starting Python (a per-session set CURIE_DATA_DIR=... in cmd, or $env:CURIE_DATA_DIR = '...' in PowerShell, requires no administrator rights). On a shared (multi-user) machine, an administrator can instead populate the site-wide data directory once — /usr/local/share/curie on Linux, /Library/Application Support/curie on macOS, C:\ProgramData\curie on Windows — and Curie uses those files read-only in place from every account, with nothing downloaded or duplicated per user. Files placed manually (in the data directory, CURIE_DATA_DIR, or the site-wide directory) are trusted as provided: keep them in sync with the data release matching the installed Curie version when upgrading.