Generate the estimator tables from one source

[xhan97]

The tables listing what IKPyKit implements existed in three places: the root README.md, docs/README.md, and — since the API reference became generated — api/index.md. The two READMEs share 120 identical lines.

The drift this was hiding

Comparing the copies turned up four errors that had been sitting there:

IForest publication	The summary matrix said TKDD'12, the table right below it said TKDD2022. On the same page. The paper is TKDD 2012, so the matrix was right and the table wrong.
STREAMKHC description	"Isolation Distribution Kernel for Trajectory Anomaly Detections" — that is IKAT's description, copy-pasted. It is a streaming hierarchical clustering algorithm.
TIDKC	"Distribution-based Tajectory Clustering"
Names	The matrix called IsoDisKernel "IsodisKernel", and the class STREAMKHC "StreaKHC".

What can and cannot be generated

Only half of each table is in the code:

Column	Source
Abbr + link	__all__ of each module — derived
Algorithm	Often the paper title, not the docstring summary — not derivable
Application	Not in the code
Publication	The venue is in the docstring References section, but as free-form prose ("In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33, 2019, July, pp. 4755-4762" → AAAI2019). Parsing that would be fragile.

So scripts/estimators.py holds what the code does not state, one entry per estimator, and renders all the tables from it. Three copies become one. Not zero maintenance, but adding an estimator is now one entry instead of three tables.

check_complete() fails the build when a public class has no entry, or an entry has no class:

No entry in ESTIMATORS for INNE. Add one in scripts/estimators.py so the tables include it.

That is the part the old setup lacked entirely — a missing estimator just silently wasn't in the tables.

Keeping the root README in step

The root README never passes through MkDocs, so no plugin can reach it. Both READMEs carry the tables between markers, filled in by scripts/sync_readme_tables.py.

The lint job runs it with --check — that is the real gate, since the pre-commit hook only helps people who installed it. Both are wired up.

The comparison ignores padding around table pipes, so an editor or formatter realigning a table does not read as drift and start a rewrite war with the generator.

Other content changes

Single-sourcing means everything now agrees, which moves some things:

• Sections use the names the docs navigation already uses — Point Anomaly Detection not Point Anomaly detection, Graph Mining not Graph Data.
• Rows follow __all__ order, matching the API reference.
• Estimators sharing a summary-matrix cell are joined with <br> instead of spilling into blank rows with empty leading cells.
• The API overview gained a publication column from the same data.
• Roman numerals are consistently lowercase (they ran i, ii, iii, IV, V, VI, VII, VIII).

Verification

Four behaviours tested directly:

Test	Result
--check when in sync	exit 0
--check after a hand edit to a table	exit 1, names the file
--check after re-aligning table pipes	exit 0 (padding tolerated)
--check with an estimator removed from ESTIMATORS	exit 1, names the class

mkdocs build --strict passes — which also validates that every relative link in the docs README resolves to a generated API page. All pre-commit hooks pass.

🤖 Generated with Claude Code