diff options
| author | Douglas Rumbaugh <dbr4@psu.edu> | 2023-07-25 11:17:36 -0400 |
|---|---|---|
| committer | Douglas Rumbaugh <dbr4@psu.edu> | 2023-07-25 11:17:36 -0400 |
| commit | 37434f5baf632e839dc14b3c7d8745287cb9368a (patch) | |
| tree | 4b9a77c25b734872a1b815cc7c0bad6258784601 /benchmarks/vptree_knn_bench.cpp | |
| parent | 9e869d32344d5bd8ee703a0733d80d48d458217c (diff) | |
| download | dynamic-extension-37434f5baf632e839dc14b3c7d8745287cb9368a.tar.gz | |
Benchmarks: mtree and vptree benchmark updates
Note: cosine similarity doesn't seem to work for VPTree--I don't think
that it is actually a metric, upon further research. At the very least I
can't find anyone claiming it is, and I've found several people claiming
it isn't. On testing with the Word2Vec data, Euclidean distance works
insofar as the M-Tree and VPTree return the same KNN results for test
queries, whereas Cosine Similarity does not work.
Diffstat (limited to 'benchmarks/vptree_knn_bench.cpp')
| -rw-r--r-- | benchmarks/vptree_knn_bench.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/benchmarks/vptree_knn_bench.cpp b/benchmarks/vptree_knn_bench.cpp index a5c45f4..0021c4a 100644 --- a/benchmarks/vptree_knn_bench.cpp +++ b/benchmarks/vptree_knn_bench.cpp @@ -19,7 +19,7 @@ int main(int argc, char **argv) double insert_batch = 0.1; init_bench_env(record_count, true); - auto queries = read_knn_queries<de::KNNQueryParms<Word2VecRec>>(qfilename, 50); + auto queries = read_knn_queries<de::KNNQueryParms<Word2VecRec>>(qfilename, 10); auto de_vp_knn = ExtendedVPTree_KNN(buffer_cap, scale_factor, max_delete_prop); |