Skip to content

๏ฟฝ CausalIQ Analysis Project

The CausalIQ Analysis project provides tools for analysing causal discovery results, including structural evaluation metrics, graph merging, optimal DAG extraction, and metric summarisation.

Current Version: v0.4.0

Quick Links:

Key Features

๐Ÿ“ Structural Evaluation (evaluate_graph)

Compare learned graphs against ground truth using standard metrics (F1, precision, recall, SHD) and equivalence class metrics, with support for multiple graph formats. Available as CLI and workflow action.

๐ŸŽฏ Optimal DAG Extraction (best_graph)

Extract the best DAG from a Probabilistic Dependency Graph using a greedy algorithm with configurable edge probability thresholds. Available as CLI and workflow action.

๐Ÿ“Š Metric Summarisation (summarise)

Aggregate metrics across experiments into publication-ready CSV tables with configurable summary statistics and filter expressions. Available as CLI and workflow action.

๐Ÿ”— Graph Merging (merge_graphs)

Combine multiple learned graphs into a single PDG using average, noisy-OR, or max strategies with optional metadata-driven weights. Available as CLI and workflow action.

๐Ÿ”„ Trace Migration (migrate_trace)

Convert legacy experiment traces (pickle format) to modern GraphML and JSON metadata, with filtering by sample size and seeds. Available as CLI and workflow action.

๐Ÿ“ Foundation Metrics

CausalIQ and Bayesys structural graph metrics and KL divergence metric.

Upcoming Key Innovations

๐Ÿง  LLM-assisted Graph Averaging

  • Uncertain or conflicting edges โ€” resolved using LLM queries

๐Ÿ“Š Publication-ready chart generation

  • Seaborn charts โ€” flexible, standardised publication-ready chart generation

โ–ฆ Publication-ready table generation

  • LaTeX tables โ€” converts tabular analysis data into publication-ready LaTeX tables

Integration with CausalIQ Ecosystem

  • ๐Ÿ” CausalIQ Discovery generates causal graphs which this package evaluates and visualises.
  • ๐Ÿค– CausalIQ Workflow can access all features of this package (through the Action interface) so that analysis and visualisation are incorporated into CausalIQ workflows.
  • ๐Ÿงช CausalIQ Research uses the analysis, table and chart features of this package to generate published paper assets.


CausalIQ Analysis provides the analytical foundation for evaluating, comparing, and summarising causal discovery results across the entire CausalIQ ecosystem.