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 +1,64 @@
SELECT hamming_distance(B'111', B'111');
hamming_distance
------------------
0
(1 row)
SELECT hamming_distance(B'111', B'110');
hamming_distance
------------------
1
(1 row)
SELECT hamming_distance(B'111', B'100');
hamming_distance
------------------
2
(1 row)
SELECT hamming_distance(B'111', B'000');
hamming_distance
------------------
3
(1 row)
SELECT hamming_distance(B'111', B'00');
ERROR: different bit lengths 3 and 2
SELECT jaccard_distance(B'1111', B'1111');
jaccard_distance
------------------
0
(1 row)
SELECT jaccard_distance(B'1111', B'1110');
jaccard_distance
------------------
0.25
(1 row)
SELECT jaccard_distance(B'1111', B'1100');
jaccard_distance
------------------
0.5
(1 row)
SELECT jaccard_distance(B'1111', B'1000');
jaccard_distance
------------------
0.75
(1 row)
SELECT jaccard_distance(B'1111', B'0000');
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance(B'1100', B'1000');
jaccard_distance
------------------
0.5
(1 row)
SELECT jaccard_distance(B'1111', B'000');
ERROR: different bit lengths 4 and 3

View File

@@ -208,6 +208,18 @@ SELECT l1_distance('[3e38]'::vector, '[-3e38]');
Infinity
(1 row)
SELECT quantize_binary('[1,0,-1]');
quantize_binary
-----------------
100
(1 row)
SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]');
quantize_binary
-----------------
01001110101
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
avg
-----------

View File

@@ -0,0 +1,21 @@
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';
val
-----
111
110
100
000
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
count
-------
4
(1 row)
DROP TABLE t;

View File

@@ -0,0 +1,21 @@
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';
val
------
1111
1110
1100
0000
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
count
-------
4
(1 row)
DROP TABLE t;