Jorge Alonso Ortiz

ON YOUR COMPUTER

One pip install

The base install pulls seven dependencies and covers the estimators, the data fetchers and the parquet code paths. The first check runs offline, with no data and no API key.

pip install puremacro
python -m puremacro.examples.sign_restrictions_uhlig

Optional extras: [backend] (numba, Apple-Silicon mlx), [cuda], [data], [narrative], [local-llm], [notebooks].

IN YOUR BROWSER

A playground with nothing to install

The JupyterLite playground runs puremacro on a Pyodide kernel inside the browser tab, with 33 notebooks and their Spanish twins.

ON AN IPAD

Pyodide and juno.sh

Install the estimator core without dependency resolution, then ask the runtime what the device can actually do.

import micropip
await micropip.install("puremacro", deps=False)
await micropip.install(["numpy", "scipy", "pandas", "matplotlib", "requests"])

from puremacro import runtime
print(runtime.report())
#  host       : pyodide 3.12.7 (wasm32)
#  device     : tablet
#  network    : js-fetch (call runtime.enable_browser_network())
#  parquet    : unavailable -> use puremacro.runtime.store / pocket

WHEN THE TABLET IS NOT ENOUGH

Offload to Google Colab

Package a heavy job, such as a 10,000-draw MCMC or a large bootstrap, as a self-contained notebook, run it on Colab and load the result back as a portable cartridge.

from puremacro.runtime.colab import generate_colab_notebook, load_colab_result

generate_colab_notebook(
    """
import puremacro as pm
result = pm.dsge.estimate_sw07(n_draws=10000, n_chains=4)
""",
    mount_drive=True,
    save_path="sw07_offload.ipynb",
    output_filename="sw07_posterior.pmz",
)
# posterior = load_colab_result("sw07_posterior.pmz")