Graph projections and catalogue

Wiki / Graph data science

On this page

• Graph catalogue
  • Native projection estimates

Graph catalogue

These statements illustrate catalogue operations on existing Person/KNOWS data. Run the self-contained walkthrough for a fresh database. Keep the projection until you have finished the algorithm examples.

CALL gds.graph.project('people', 'Person', {
  KNOWS: {orientation: 'UNDIRECTED', properties: ['weight']}
}, {nodeProperties: ['embedding']});
CALL gds.graph.list() YIELD graphName, nodeCount, relationshipCount;
CALL gds.graph.exists('people') YIELD exists;

project(name, nodeProjection, relationshipProjection[, configuration]) accepts a string, a list of strings or a map for either projection. '*' selects all labels/types. Map entries can rename a label/type using label or type; relationships support NATURAL, REVERSE and UNDIRECTED orientation. Undirected projections contain an arc in each direction. Both map forms accept properties; global nodeProperties and relationshipProperties are also supported. Property definitions may rename fields and provide defaults:

CALL gds.graph.project('weighted', '*', {
  KNOWS: {properties: {weight: {property: 'cost', defaultValue: 1.0}}}
});

Projected graphs are immutable snapshots until explicitly mutated. They belong to one database, remain independent of subsequent live graph changes, consume RAM and disappear on reopen. Project/drop/mutate participate in GDB statement and explicit transaction rollback. Failed write-back rejects deleted nodes and reused numeric node IDs. Reproject to analyse newly created data.

list([name]) returns catalogue entries, exists(name) returns a boolean, and drop(name[, failIfMissing]) defaults to failing on an unknown graph. For example, CALL gds.graph.drop('people') releases this projection after use. gds.graph.nodeProperty.stream(name, property) returns nodeId, propertyValue. gds.graph.nodeProperties.write(name, properties) copies projected properties onto live nodes; properties can be a name or list of names. gds.util.asNode(id) and gds.util.asNodes(ids) resolve live node values.

Native projection is implemented; Cypher aggregation/query projections, catalogue persistence and model/pipeline catalogues are not included.

Native projection estimates

CALL gds.graph.project.estimate('Person', {
  KNOWS: {orientation:'UNDIRECTED', properties:['weight']}
}, {nodeProperties:['embedding']})
YIELD nodeCount, relationshipCount, bytesMin, bytesMax, requiredMemory;

The arguments are node projection, relationship projection and optional configuration; no graph name is supplied. This read-only scan uses the same selection/property rules as native projection, including aliasing and doubled undirected arcs. It retains selected IDs but does not create a catalogue entry or allocate a complete projected graph. Memory ranges are planning estimates, not allocator measurements or reservations.

Related articles

GDS algorithms and execution modes · GDS resources and cancellation · Graph data science