Closed mneedham closed 5 years ago
Simple benchmark with a dummy compute doesn't show much difference in the time taken to compute all pairs:
Benchmark (concurrency) Mode Cnt Score Error Units
SimilarityStreamGeneratorBenchmark.allPairs 1 ss 1000 138249.941 ± 16260.097 ns/op
SimilarityStreamGeneratorBenchmark.allPairs 2 ss 1000 22361607.769 ± 2862052.577 ns/op
SimilarityStreamGeneratorBenchmark.allPairs 8 ss 1000 19253922.198 ± 2185645.041 ns/op
SimilarityStreamGeneratorBenchmark.allPairsBlankSourceTarget 1 ss 1000 82898.555 ± 2002.081 ns/op
SimilarityStreamGeneratorBenchmark.allPairsBlankSourceTarget 2 ss 1000 18480648.936 ± 2548120.298 ns/op
SimilarityStreamGeneratorBenchmark.allPairsBlankSourceTarget 8 ss 1000 20026972.757 ± 2267054.050 ns/op
Will also test this out on the recipes dataset.
This PR adds functionality discussed on the Neo4j community site - https://community.neo4j.com/t/algo-similarity-jaccard-stream-takes-more-than-3-minutes/4586/14
Things still to do: