ECG

SQL function: cugraph_ecg

Official cuGraph reference: C API

Stabilize community assignments by combining an ensemble of randomized Louvain partitions into a final consensus clustering.

Signature

cugraph_ecg(table_name [, src_col, dst_col [, weight_col [, options_json]]])

Relation inputs

The first positional argument names a registered edge table or view (the edges role). Parenthesized relation subqueries are not accepted; metadata validation uses the same registered name.

Vertex ID types

The edges relation declares the accepted vertex-ID domains. Numeric calls preserve the existing numeric schema. When logical string support is declared, Utf8, LargeUtf8, and Utf8View endpoint columns share one logical domain; their vertex-identity outputs are canonicalized to Utf8.

Domain	Accepted endpoint inputs	Output contract
Numeric edge endpoints	Int32, Int64	The numeric output schema is used for numeric calls.
Logical string edge endpoints	Utf8, LargeUtf8, Utf8View	Vertex identity columns are canonicalized to Utf8; scores, distances, counts, coordinates, and opaque labels remain numeric.

The native mapping type is Int64. Call-specific output schemas come from gpu_validate_call.

Logical string side-input limitations:

• edge ID columns and edge-ID predicate side inputs are not supported for logical string graphs

Scalar arguments & JSON options

Positional scalar arguments

src_col and dst_col name the edge endpoint columns; both are optional and default to src and dst.

Argument	Type	Required	Default	Notes
weight_col	Utf8|null	no		accepted as an edge-column binding; native algorithm execution does not consume weights; semantic effect: none for this algorithm

JSON options

Option	Type	Default	Constraints	Description
ensemble_size	UInt32	10	min 1
max_level	UInt32	100	min 1
min_weight	Float64	0.001	min 0
resolution	Float64	1	> 0
seed	UInt64	0
threshold	Float64	1e-7	min 0

Graph construction options

This function builds an undirected graph by default (directed=false); all other graph construction options follow the shared defaults documented in Graph Construction Options.

Output schema

Column	Type	Nullable	Description
vertex	Int64|Utf8	no	Vertex assigned to an ECG community.
partition	Int64	no	Community identifier assigned by ECG.

These are generic descriptor schemas; validate the call to get the concrete, table-specific output schema.

Examples

This example runs on the citation network demo dataset.

Benchmark the ensemble against single-run algorithms

ECG runs an ensemble of Louvain passes (ensemble_size, default 10), reweights edges by how often their endpoints co-cluster, and clusters the consensus. Is that worth it? Because every cugraph_* function returns a plain relation, one statement can run all three community algorithms on the same subgraph (the Louvain example's 2010s AI views) and score them against the human-assigned primary_fos labels — the share of members in a community carrying its dominant label:

CREATE OR REPLACE VIEW ai_nodes AS
SELECT paper_id FROM papers
WHERE year >= 2010 AND primary_fos IN (
  'Deep learning', 'Artificial neural network', 'Convolutional neural network',
  'Recurrent neural network', 'Natural language processing',
  'Reinforcement learning', 'Image segmentation', 'Feature extraction',
  'Object detection', 'Speech recognition');

CREATE OR REPLACE VIEW ai_edges AS
SELECT e.src, e.dst
FROM citation_edges e
JOIN ai_nodes a ON a.paper_id = e.src
JOIN ai_nodes b ON b.paper_id = e.dst;

WITH labeled AS (
  SELECT 'louvain' AS algorithm, c."partition" AS community, p.primary_fos
  FROM cugraph_louvain('ai_edges', 'src', 'dst') c
  JOIN papers p ON p.paper_id = c.vertex
  UNION ALL
  SELECT 'leiden', c."partition", p.primary_fos
  FROM cugraph_leiden('ai_edges', 'src', 'dst') c
  JOIN papers p ON p.paper_id = c.vertex
  UNION ALL
  SELECT 'ecg', c."partition", p.primary_fos
  FROM cugraph_ecg('ai_edges', 'src', 'dst') c
  JOIN papers p ON p.paper_id = c.vertex),
counts AS (
  SELECT algorithm, community, primary_fos, COUNT(*) AS n
  FROM labeled GROUP BY 1, 2, 3),
sized AS (
  SELECT algorithm, community, SUM(n) AS members, MAX(n) AS top_label
  FROM counts GROUP BY 1, 2)
SELECT algorithm,
       COUNT(*) AS communities,
       COUNT(*) FILTER (WHERE members >= 100) AS ge100,
       ROUND(SUM(top_label) FILTER (WHERE members >= 100) * 100.0
             / SUM(members) FILTER (WHERE members >= 100), 1) AS purity_pct
FROM sized
GROUP BY algorithm
ORDER BY purity_pct DESC;

algorithm	communities	ge100	purity_pct
leiden	2,670	16	44.1
ecg	5,014	41	41.7
louvain	1,960	14	38.5

Three GPU graph builds plus the joins and aggregation return together in about one second. ECG's consensus is deliberately conservative: it only keeps vertices together when most ensemble members agree, so it produces the finest partition (5,014 communities, 41 of them with 100+ papers) and beats a single Louvain run on label purity. Determinism comes from seed (default 0); ensemble_size trades run time for consensus stability.

Limitations & lifecycle

No algorithm-specific limitations.

Validate before running

Dry-run validation checks registered relation metadata, column presence, static dtypes, and options only; it does not scan edge data, construct a graph, or prove source-vertex existence:

SELECT * FROM gpu_validate_call(
  'cugraph_ecg',
  '{"schema_version":1,"relations":{"edges":{"table":"target_edges"}},"options":{"src_col":"src","dst_col":"dst"}}'
);

See GPU Function Catalog API for the full gpu_validate_call contract.