cancel
Showing results for 
Search instead for 
Did you mean: 

About projectedFeatureDimension in graphSage model

lingvisa
Graph Fellow

In today's seminar, the MS News graphSage model configuration is below:

CALL gds.beta.graphSage.train(
  'mind-gs-projection',
  {
    modelName: 'graphSage',
    featureProperties: ['wikiEncoding'],
    aggregator: 'mean',
    activationFunction: 'sigmoid',
    projectedFeatureDimension: 10,
    sampleSizes: [25, 10],
    tolerance: 0.001,
    searchDepth: 5,
    embeddingDimension: 128
  }
)

The data preprocessing creates entity embedding of 100 dimension. And according the documentation:
The projectedFeatureDimensionequals the maximum length of the feature-array, i.e.,ageandcostboth are scalar features plus the list featureheightAndWeightwhich has a length of two. For each node its unique labels properties is projected using a label specific projection to vector space of dimensionprojectedFeatureDimension .

So the projectedFeatureDimension should be at least 100 in the cypher command above. But why is it only 10? Please clarify the proper choice of this parameter.

1 REPLY 1

lingvisa
Graph Fellow

Please take a look at this question.

Nodes 2022
Nodes
NODES 2022, Neo4j Online Education Summit

On November 16 and 17 for 24 hours across all timezones, you’ll learn about best practices for beginners and experts alike.