Skip to content

CausalIQ Analysis - Development Roadmap

Last updated: November 09, 2026

This project roadmap fits into the overall ecosystem roadmap.

🚧 Under Development [October 2026]

v0.5.0 — Reproduce variable-ordering paper

Assuring reproducibility of results and assets for published papers.

  1. ✅ workflow cache can be used as reference for evaluate_graph
  2. ✅ new plot action works in GitHub CI with no warnings
  3. ✅ Support tuples, lists and dicts as property values
  4. ✅ new edge and equiv.edge metrics for evaluate_graph return the ten edge comparison counts
  5. ✅ develop pdg_compare based on 9 edge comparison counts
  6. ✅ Diagnose failing legacy tests
  7. ✅ Replace deprecated matplotlib warnings (cap pyparsing<3.3)
  8. ✅ Ensure all modules meet development guidelines
  9. ✅ Remove UPDATE & NOCACHES and add PRODUCE and MAP Action Patterns1.
  10. ✅ Move to V2 metedata structure
  11. ✅ Ensure all modules meet development guidelines
  12. ✅ Summarise a table from multiple input caches, in CSV, LaTeX or Markdown
  13. ✅ Remove object_type parameter, replaced by the cpdag flag
  14. 📊 Support Python 3.10-14, with 3.12 as default, and badge as SPEC 0 — Minimum Supported Dependencies
  15. 🛣️ When Python 3.9 support is dropped, move to matplotlib >=3.10.7 (removes the pyparsing cap; requires re-blessing the ord_hc_f1 SVG reference)
  16. 🛣️ Optimise summarise (aggregate workflow patterns) so it is much faster

✅ Implemented Releases

v0.4.0 — Evaluation and Summarisation

  • Structural Evaluation (evaluate_graph) — Compute F1, precision, recall, SHD vs ground truth graphs, including equivalence class metrics
  • Optimal DAG Extraction (best_graph) — Extract optimal DAG from PDGs using greedy algorithm with cycle avoidance
  • Summarisation (summarise) — Summarise numerical metrics across experiments into publication-ready statistics (mean, SD, count) with CSV output
  • Merge Strategies — Added noisy_or and max merge strategies
  • Validation Module — Shared utilities for parameter validation, seed ranges, and filter expressions

v0.3.0 — Graph Merging

Workflow actions to migrate legacy traces and merge graphs.

  • Trace Migration (migrate_trace) — Convert legacy Trace pickle files to DAG/CPDAG in workflow caches
  • Graph Merging (merge_graphs) — Merge multiple DAGs/CPDAGs/PDGs into a single PDG with edge probabilities, supporting metadata-driven weighting

v0.2.0 — Legacy Trace Support

  • Support for reading and writing structure learning traces in legacy pickle format

v0.1.0 — Foundation Metrics

  • CausalIQ and Bayesys structural graph metrics
  • KL divergence metric

See Git commit history for detailed implementation progress.

Documentation