Added support for bit vectors to HNSW

This commit is contained in:
Andrew Kane
2024-04-01 20:30:55 -07:00
parent 7ee9074a9c
commit 94a444f029
21 changed files with 541 additions and 5 deletions

View File

@@ -0,0 +1,13 @@
SELECT hamming_distance(B'111', B'111');
SELECT hamming_distance(B'111', B'110');
SELECT hamming_distance(B'111', B'100');
SELECT hamming_distance(B'111', B'000');
SELECT hamming_distance(B'111', B'00');
SELECT jaccard_distance(B'1111', B'1111');
SELECT jaccard_distance(B'1111', B'1110');
SELECT jaccard_distance(B'1111', B'1100');
SELECT jaccard_distance(B'1111', B'1000');
SELECT jaccard_distance(B'1111', B'0000');
SELECT jaccard_distance(B'1100', B'1000');
SELECT jaccard_distance(B'1111', B'000');

View File

@@ -48,6 +48,9 @@ SELECT l1_distance('[0,0]'::vector, '[0,1]');
SELECT l1_distance('[1,2]'::vector, '[3]');
SELECT l1_distance('[3e38]'::vector, '[-3e38]');
SELECT quantize_binary('[1,0,-1]');
SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]');
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;

12
test/sql/hnsw_hamming.sql Normal file
View File

@@ -0,0 +1,12 @@
SET enable_seqscan = off;
CREATE TABLE t (val bit(3));
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
INSERT INTO t (val) VALUES (B'110');
SELECT * FROM t ORDER BY val <~> B'111';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
DROP TABLE t;

12
test/sql/hnsw_jaccard.sql Normal file
View File

@@ -0,0 +1,12 @@
SET enable_seqscan = off;
CREATE TABLE t (val bit(4));
INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
INSERT INTO t (val) VALUES (B'1110');
SELECT * FROM t ORDER BY val <%> B'1111';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
DROP TABLE t;