Install & set up¶

Both libraries live in the intelligentactuaries/scelo repository, in packages/scelo-py and packages/scelo-r, and install from a checkout in one command each. They are deliberately light: the Python core needs numpy and pandas alone; the R package needs base R alone.

Python¶

pip install scelo

scelo on PyPI. Python 3.9+ on any OS. Two other routes to the same thing:

# straight from GitHub
pip install "scelo @ git+https://github.com/intelligentactuaries/scelo#subdirectory=packages/scelo-py"

# from a repository checkout
git clone https://github.com/intelligentactuaries/scelo
pip install scelo/packages/scelo-py

Extras¶

The core installs numpy + pandas only. Everything statistical has a pure numpy implementation, and upgrades itself when the optional packages are present:

Extra	Adds	Unlocks
`scelo[stats]`	scipy, statsmodels	GLMs via statsmodels (the numpy GLM agrees to 1e-5), scipy distribution fits
`scelo[life]`	lifelib 0.14.0, modelx 0.32.0, openpyxl	`sc.lifelib_run()` — the real lifelib models, the pair Scelo IDE ships
`scelo[reserving]`	chainladder	cross-checks against the chainladder package
`scelo[io]`	pyarrow, openpyxl	parquet and Excel in `sc.load()` / `sc.export()`
`scelo[viz]`	matplotlib	the `sc.plot_*` chart family
`scelo[all]`	all of the above
`scelo[dev]`	pytest + the working set	`pytest` runs the golden-value suite

pip install "scelo[stats,viz]"

R¶

install.packages("scelo", repos = c("https://intelligentactuaries.com/r", getOption("repos")))
library(scelo)

That is the lab's own CRAN-style repository at intelligentactuaries.com/r — CRAN itself requires an open-source licence, so the package is distributed from there and from GitHub instead. The package is pure R, so the source install works on Windows, macOS and Linux without build tools. Two other routes to the same thing:

# straight from GitHub
install.packages("remotes")
remotes::install_github("intelligentactuaries/scelo", subdir = "packages/scelo-r")

# from a repository checkout
install.packages("scelo/packages/scelo-r", repos = NULL, type = "source")

R ≥ 4.1. The package imports nothing beyond stats, utils and tools, so it installs on a bare R. Optional packages add extras rather than gate the basics:

Package	Unlocks
`jsonlite`, `curl`	the swarm client (`sc_council`, `sc_society`, `sc_augment`)
`statmod`	Tweedie GLMs in `sc_glm()`
`reticulate`	`sc_lifelib_run()` — lifelib models through Python
`testthat`	`tests/` — the parity suite against the Python goldens

Inside Scelo IDE¶

Scelo IDE bundles its own CPython and R. The libraries run on both as they are: open the IDE terminal and run the same pip install / install.packages() lines against the bundled runtimes. Everything the libraries compute matches what the IDE's own panels compute, down to the random stream — that is the point of them.

The swarm (optional)¶

sc.wmtr / sc_wmtr need no server: the W(M, T, R) engine is ported into each library. The deliberation functions — sc.council, sc.society, sc.augment and friends — talk to the Scelo swarm, a local Bun server:

Scelo IDE 0.1.6+ starts it automatically on 127.0.0.1:3010 while the app is open — nothing to do.
From a repository checkout, bun run dev:swarm starts the same server.
sc.connect("http://host:3010") / sc_connect("http://host:3010") points the client somewhere else.

Without a reachable swarm those functions say so and stop; nothing else in the libraries depends on it.

Check the install¶

import scelo as sc
sc.life_table().head(3)     # prints the table, its basis and its caveats
sc.cheatsheet()             # the one-screen map of everything

library(scelo)
sc_life_table()[1:3, ]
sc_cheatsheet()

If the life table prints with its basis line and notes underneath, the library is working. The illustrative Gompertz–Makeham warning in those notes is not an error — it is the library telling you, as it always will, what basis produced the numbers you are looking at.