Skip to content

Foundations

Every pillar in ergodic reads the same two things: a causal graph and the data. Two more objects sit between them, the domain knowledge that constrains the graph and the identification step that turns a graph into an estimand. These four are built first, and the rest of the library is a consumer of them.

  • Causal graphs

    One mixed-graph object covers DAG, ADMG, CPDAG, MAG, and PAG. Edges read as glyphs like X -> Y, and graphs are immutable values: every edit returns a new graph.

    Causal graphs

  • Domain knowledge

    Prior constraints that discovery and inference read: required and forbidden edges, tiers, roots, sinks, confounded pairs. Immutable, and composable with &.

    Domain knowledge

  • Data

    Typed datasets across five shapes: tabular, time series, panel, hierarchical, and event log. A dataframe builds one and stays at the boundary; the math runs in numpy.

    Data

  • Identification

    From a graph to an estimand, on a single graph or a whole equivalence class. Adjustment, front-door, instruments, and the general do-calculus ID algorithm.

    Identification

New to the ideas behind these objects? The concepts page covers the model behind the graph and the knowledge set in plain words, so the method-by-method guides read easily.