ergodic

Causal AI in one Python package. ergodic learns causal graphs from data, tells you whether a question can be answered from them, and estimates the effect with an interval. It also covers quasi-experiments on panels, process mining on event logs, structural causal models, and Bayesian forecasting with drivers. ergodic runs locally; the optional [welt] extra calls the hosted Welt service. ergodic is pre-1.0, so the API can change between minor versions.

Documentation: docs.ergodic.dev

Install

pip install "ergodic"
# or
uv add "ergodic"

Requires Python 3.11 or later. The extras "ergodic[scm]", "ergodic[forecast]" and "ergodic[latent]" add PyMC for structural causal models, forecasting and hidden regimes. "ergodic[welt]" connects to Welt, which runs hosted models on your data.

Your first causal effect

from ergodic import discover, estimate_effect, identify_effect
from ergodic.datasets import load_earthquake

alarm = load_earthquake(n=10_000, seed=0)  # the burglar-alarm network (Pearl, 1988)
graph = discover(alarm.data, method="pc").graph  # learn the graph from data alone
for edge in sorted(edge.glyph() for edge in graph.edges):
    print(edge)

print(identify_effect(graph, "Burglary", "JohnCalls"))  # can the data answer it?
print(estimate_effect(graph, alarm.data, "Burglary", "JohnCalls", treated="True"))

Alarm --> JohnCalls
Alarm --> MaryCalls
Alarm <-- Burglary
Alarm <-- Earthquake
Identification(adjustment: P(JohnCalls | do(Burglary)) = sum_z P(JohnCalls | Burglary, z) P(z), with z = {Earthquake})
EffectEstimate(aipw, analytic: ate=0.7982, se=0.0333, 95% CI [0.733, 0.863], n=10000)

The true effect, from the network's own tables, is 0.793: inside the interval. The documentation starts from this program and goes step by step.

Links

• Documentation
• Welt: hosted models for prediction, causal discovery and forecasting
• Changelog, contributing, security

Licence

Apache-2.0. See LICENSE and NOTICE. Made by Ergodic.