liuyu/pgvector
1
0
Fork 0
mirror of https://github.com/pgvector/pgvector.git synced 2026-07-01 02:02:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

Removed dimensions from sparsevec

Browse Source
This commit is contained in:
Andrew Kane
2024-04-03 20:55:03 -07:00
parent 35d0fe88b9
commit dde3a2aacd
17 changed files with 166 additions and 289 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sql/vector--0.6.2--0.7.0.sql
View File

@@ -173,9 +173,6 @@ CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -185,7 +182,6 @@ CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
CREATE TYPE sparsevec (
INPUT = sparsevec_in,
OUTPUT = sparsevec_out,
TYPMOD_IN = sparsevec_typmod_in,
RECEIVE = sparsevec_recv,
SEND = sparsevec_send,
STORAGE = external

4
sql/vector.sql
View File

@@ -480,9 +480,6 @@ CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -492,7 +489,6 @@ CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
CREATE TYPE sparsevec (
INPUT = sparsevec_in,
OUTPUT = sparsevec_out,
TYPMOD_IN = sparsevec_typmod_in,
RECEIVE = sparsevec_recv,
SEND = sparsevec_send,
STORAGE = external

19
src/hnswbuild.c
View File

@@ -681,8 +681,6 @@ GetMaxDimensions(HnswType type)
maxDimensions *= 2;
else if (type == HNSW_TYPE_BIT)
maxDimensions *= 32;
else if (type == HNSW_TYPE_SPARSEVEC)
maxDimensions = INT_MAX;
return maxDimensions;
}
@@ -693,8 +691,6 @@ GetMaxDimensions(HnswType type)
static void
InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
{
int maxDimensions;
buildstate->heap = heap;
buildstate->index = index;
buildstate->indexInfo = indexInfo;
@@ -705,14 +701,17 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
buildstate->efConstruction = HnswGetEfConstruction(index);
buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
maxDimensions = GetMaxDimensions(buildstate->type);
/* Require column to have dimensions to be indexed */
if (buildstate->dimensions < 0)
elog(ERROR, "column does not have dimensions");
if (buildstate->type != HNSW_TYPE_SPARSEVEC)
{
int maxDimensions = GetMaxDimensions(buildstate->type);
if (buildstate->dimensions > maxDimensions)
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
if (buildstate->dimensions < 0)
elog(ERROR, "column does not have dimensions");
if (buildstate->dimensions > maxDimensions)
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
}
if (buildstate->efConstruction < 2 * buildstate->m)
elog(ERROR, "ef_construction must be greater than or equal to 2 * m");

2
src/hnswutils.c
View File

@@ -229,7 +229,7 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
else if (type == HNSW_TYPE_SPARSEVEC)
{
SparseVector *v = DatumGetSparseVector(*value);
SparseVector *result = InitSparseVector(v->dim, v->nnz);
SparseVector *result = InitSparseVector(v->nnz);
float *vx = SPARSEVEC_VALUES(v);
float *rx = SPARSEVEC_VALUES(result);

168
src/sparsevec.c
View File

@@ -18,69 +18,28 @@
#include "utils/builtins.h"
#endif
/*
* Ensure same dimensions
*/
static inline void
CheckDims(SparseVector * a, SparseVector * b)
{
if (a->dim != b->dim)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("different sparsevec dimensions %d and %d", a->dim, b->dim)));
}
/*
* Ensure expected dimensions
*/
static inline void
CheckExpectedDim(int32 typmod, int dim)
{
if (typmod != -1 && typmod != dim)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("expected %d dimensions, not %d", typmod, dim)));
}
/*
* Ensure valid dimensions
*/
static inline void
CheckDim(int dim)
{
if (dim < 1)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("sparsevec must have at least 1 dimension")));
if (dim > SPARSEVEC_MAX_DIM)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("sparsevec cannot have more than %d dimensions", SPARSEVEC_MAX_DIM)));
}
/*
* Ensure valid nnz
*/
static inline void
CheckNnz(int nnz, int dim)
CheckNnz(int nnz)
{
if (nnz < 0)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("sparsevec must have at least one element")));
if (nnz > dim)
if (nnz > SPARSEVEC_MAX_NNZ)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("sparsevec cannot have more elements than dimensions")));
errmsg("sparsevec cannot have more elements than non-zero elements")));
}
/*
* Ensure valid index
*/
static inline void
CheckIndex(int32 *indices, int i, int dim)
CheckIndex(int32 *indices, int i)
{
int32 index = indices[i];
@@ -89,11 +48,6 @@ CheckIndex(int32 *indices, int i, int dim)
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("index must not be negative")));
if (index >= dim)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("index must be less than dimensions")));
if (i > 0)
{
if (index < indices[i - 1])
@@ -129,7 +83,7 @@ CheckElement(float value)
* Allocate and initialize a new sparse vector
*/
SparseVector *
InitSparseVector(int dim, int nnz)
InitSparseVector(int nnz)
{
SparseVector *result;
int size;
@@ -137,7 +91,6 @@ InitSparseVector(int dim, int nnz)
size = SPARSEVEC_SIZE(nnz);
result = (SparseVector *) palloc0(size);
SET_VARSIZE(result, size);
result->dim = dim;
result->nnz = nnz;
return result;
@@ -167,8 +120,6 @@ Datum
sparsevec_in(PG_FUNCTION_ARGS)
{
char *lit = PG_GETARG_CSTRING(0);
int32 typmod = PG_GETARG_INT32(2);
int dim;
char *pt;
char *stringEnd;
SparseVector *result;
@@ -190,6 +141,11 @@ sparsevec_in(PG_FUNCTION_ARGS)
pt++;
}
if (maxNnz > SPARSEVEC_MAX_NNZ)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("sparsevec cannot have more than %d non-zero elements", SPARSEVEC_MAX_NNZ)));
indices = palloc(maxNnz * sizeof(int32));
values = palloc(maxNnz * sizeof(float));
@@ -297,24 +253,6 @@ sparsevec_in(PG_FUNCTION_ARGS)
stringEnd++;
if (*stringEnd != '/')
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed sparsevec literal: \"%s\"", lit),
errdetail("Unexpected end of input.")));
stringEnd++;
/* Use similar logic as int2vectorin */
errno = 0;
pt = stringEnd;
dim = strtol(pt, &stringEnd, 10);
if (stringEnd == pt)
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
/* Only whitespace is allowed after the closing brace */
while (sparsevec_isspace(*stringEnd))
stringEnd++;
@@ -323,21 +261,18 @@ sparsevec_in(PG_FUNCTION_ARGS)
ereport(ERROR,
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
errmsg("malformed sparsevec literal: \"%s\"", lit),
errdetail("Junk after closing.")));
errdetail("Junk after closing right brace.")));
pfree(litcopy);
CheckDim(dim);
CheckExpectedDim(typmod, dim);
result = InitSparseVector(dim, nnz);
result = InitSparseVector(nnz);
rvalues = SPARSEVEC_VALUES(result);
for (int i = 0; i < nnz; i++)
{
result->indices[i] = indices[i];
rvalues[i] = values[i];
CheckIndex(result->indices, i, dim);
CheckIndex(result->indices, i);
CheckElement(rvalues[i]);
}
@@ -382,11 +317,9 @@ sparsevec_out(PG_FUNCTION_ARGS)
*
* nnz - 1 bytes for ,
*
* 10 bytes for dimensions
*
* 4 bytes for {, }, /, and \0
* 3 bytes for {, }, and \0
*/
buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 13);
buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 2);
ptr = buf;
AppendChar(ptr, '{');
@@ -402,45 +335,12 @@ sparsevec_out(PG_FUNCTION_ARGS)
}
AppendChar(ptr, '}');
AppendChar(ptr, '/');
AppendInt(ptr, sparsevec->dim);
*ptr = '\0';
PG_FREE_IF_COPY(sparsevec, 0);
PG_RETURN_CSTRING(buf);
}
/*
* Convert type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_typmod_in);
Datum
sparsevec_typmod_in(PG_FUNCTION_ARGS)
{
ArrayType *ta = PG_GETARG_ARRAYTYPE_P(0);
int32 *tl;
int n;
tl = ArrayGetIntegerTypmods(ta, &n);
if (n != 1)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid type modifier")));
if (*tl < 1)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("dimensions for type sparsevec must be at least 1")));
if (*tl > SPARSEVEC_MAX_DIM)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("dimensions for type sparsevec cannot exceed %d", SPARSEVEC_MAX_DIM)));
PG_RETURN_INT32(*tl);
}
/*
* Convert external binary representation to internal representation
*/
@@ -449,33 +349,30 @@ Datum
sparsevec_recv(PG_FUNCTION_ARGS)
{
StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
int32 typmod = PG_GETARG_INT32(2);
SparseVector *result;
int32 dim;
int32 nnz;
int32 unused;
int32 unused2;
float *values;
dim = pq_getmsgint(buf, sizeof(int32));
nnz = pq_getmsgint(buf, sizeof(int32));
unused = pq_getmsgint(buf, sizeof(int32));
unused2 = pq_getmsgint(buf, sizeof(int32));
CheckDim(dim);
CheckNnz(nnz, dim);
CheckExpectedDim(typmod, dim);
CheckNnz(nnz);
if (unused != 0)
if (unused != 0 || unused2 != 0)
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("expected unused to be 0, not %d", unused)));
result = InitSparseVector(dim, nnz);
result = InitSparseVector(nnz);
values = SPARSEVEC_VALUES(result);
for (int i = 0; i < nnz; i++)
{
result->indices[i] = pq_getmsgint(buf, sizeof(int32));
CheckIndex(result->indices, i, dim);
CheckIndex(result->indices, i);
}
for (int i = 0; i < nnz; i++)
@@ -499,9 +396,9 @@ sparsevec_send(PG_FUNCTION_ARGS)
StringInfoData buf;
pq_begintypsend(&buf);
pq_sendint(&buf, svec->dim, sizeof(int32));
pq_sendint(&buf, svec->nnz, sizeof(int32));
pq_sendint(&buf, svec->unused, sizeof(int32));
pq_sendint(&buf, svec->unused2, sizeof(int32));
for (int i = 0; i < svec->nnz; i++)
pq_sendint(&buf, svec->indices[i], sizeof(int32));
for (int i = 0; i < svec->nnz; i++)
@@ -512,16 +409,12 @@ sparsevec_send(PG_FUNCTION_ARGS)
/*
* Convert sparse vector to sparse vector
* This is needed to check the type modifier
*/
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec);
Datum
sparsevec(PG_FUNCTION_ARGS)
{
SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
int32 typmod = PG_GETARG_INT32(1);
CheckExpectedDim(typmod, svec->dim);
PG_RETURN_POINTER(svec);
}
@@ -534,23 +427,20 @@ Datum
vector_to_sparsevec(PG_FUNCTION_ARGS)
{
Vector *vec = PG_GETARG_VECTOR_P(0);
int32 typmod = PG_GETARG_INT32(1);
SparseVector *result;
int dim = vec->dim;
int nnz = 0;
float *values;
int j = 0;
CheckDim(dim);
CheckExpectedDim(typmod, dim);
for (int i = 0; i < dim; i++)
{
if (vec->x[i] != 0)
nnz++;
}
result = InitSparseVector(dim, nnz);
CheckNnz(nnz);
result = InitSparseVector(nnz);
values = SPARSEVEC_VALUES(result);
for (int i = 0; i < dim; i++)
{
@@ -627,8 +517,6 @@ sparsevec_l2_distance(PG_FUNCTION_ARGS)
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8(sqrt(l2_distance_squared_internal(a, b)));
}
@@ -643,8 +531,6 @@ sparsevec_l2_squared_distance(PG_FUNCTION_ARGS)
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8(l2_distance_squared_internal(a, b));
}
@@ -694,8 +580,6 @@ sparsevec_inner_product(PG_FUNCTION_ARGS)
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8(inner_product_internal(a, b));
}
@@ -709,8 +593,6 @@ sparsevec_negative_inner_product(PG_FUNCTION_ARGS)
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8(-inner_product_internal(a, b));
}
@@ -729,8 +611,6 @@ sparsevec_cosine_distance(PG_FUNCTION_ARGS)
float normb = 0.0;
double similarity;
CheckDims(a, b);
similarity = inner_product_internal(a, b);
/* Auto-vectorized */

6
src/sparsevec.h
View File

@@ -1,7 +1,7 @@
#ifndef SPARSEVEC_H
#define SPARSEVEC_H
#define SPARSEVEC_MAX_DIM 100000
#define SPARSEVEC_MAX_NNZ 100000
/* Ensure values are aligned */
#define SPARSEVEC_SIZE(_nnz) (offsetof(SparseVector, indices) + MAXALIGN((_nnz) * sizeof(int32)) + (_nnz * sizeof(float)))
@@ -13,12 +13,12 @@
typedef struct SparseVector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int32 dim; /* number of dimensions */
int32 nnz;
int32 unused;
int32 unused2;
int32 indices[FLEXIBLE_ARRAY_MEMBER];
} SparseVector;
SparseVector *InitSparseVector(int dim, int nnz);
SparseVector *InitSparseVector(int nnz);
#endif

12
src/vector.c
View File

@@ -1236,11 +1236,19 @@ sparsevec_to_vector(PG_FUNCTION_ARGS)
SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
int32 typmod = PG_GETARG_INT32(1);
Vector *result;
int dim = svec->dim;
int dim;
float *values = SPARSEVEC_VALUES(svec);
int maxIndex = svec->nnz == 0 ? -1 : svec->indices[svec->nnz - 1];
if (typmod == -1)
dim = maxIndex + 1;
else
dim = typmod;
CheckDim(dim);
CheckExpectedDim(typmod, dim);
if (dim < maxIndex + 1)
elog(ERROR, "Vector must have at least %d dimensions", maxIndex + 1);
result = InitVector(dim);
for (int i = 0; i < svec->nnz; i++)

20
test/expected/hnsw_sparsevec_cosine.out
View File

@@ -1,17 +1,17 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
val
-----------------
{0:1,1:1,2:1}/3
{0:1,1:2,2:3}/3
{0:1,1:2,2:4}/3
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
val
---------------
{0:1,1:1,2:1}
{0:1,1:2,2:3}
{0:1,1:2,2:4}
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') t2;
count
-------
3

20
test/expected/hnsw_sparsevec_ip.out
View File

@@ -1,15 +1,15 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
val
-----------------
{0:1,1:2,2:4}/3
{0:1,1:2,2:3}/3
{0:1,1:1,2:1}/3
{}/3
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
val
---------------
{0:1,1:2,2:4}
{0:1,1:2,2:3}
{0:1,1:1,2:1}
{}
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;

24
test/expected/hnsw_sparsevec_l2.out
View File

@@ -1,15 +1,15 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
val
-----------------
{0:1,1:2,2:3}/3
{0:1,1:2,2:4}/3
{0:1,1:1,2:1}/3
{}/3
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
val
---------------
{0:1,1:2,2:3}
{0:1,1:2,2:4}
{0:1,1:1,2:1}
{}
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
@@ -25,14 +25,14 @@ SELECT COUNT(*) FROM t;
(1 row)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
val
-----
(0 rows)
DROP TABLE t;
-- TODO move
CREATE TABLE t (val sparsevec(1001));
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
ERROR: sparsevec cannot have more than 1000 non-zero elements for hnsw index

22
test/expected/sparsevec_functions.out
View File

@@ -1,62 +1,60 @@
SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
SELECT l2_distance('{}'::sparsevec, '{1:1}');
l2_distance
-------------
1
(1 row)
SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
SELECT '{}'::sparsevec <-> '{0:3,1:4}';
?column?
----------
5
(1 row)
SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
inner_product
---------------
10
(1 row)
SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
sparsevec_negative_inner_product
----------------------------------
-10
(1 row)
SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
SELECT cosine_distance('{}'::sparsevec, '{}');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');
ERROR: different sparsevec dimensions 2 and 3

78
test/expected/sparsevec_input.out
View File

@@ -1,62 +1,64 @@
SELECT '{0:1.5,2:3.5}/5'::sparsevec;
sparsevec
-----------------
{0:1.5,2:3.5}/5
SELECT '{0:1.5,2:3.5}'::sparsevec;
sparsevec
---------------
{0:1.5,2:3.5}
(1 row)
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
vector
-------------
[1.5,0,3.5]
(1 row)
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
vector
-----------------
[1.5,0,3.5,0,0]
(1 row)
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
vector
-----------------
[1.5,0,3.5,0,0]
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
vector
---------------
[1.5,0,3.5,0]
(1 row)
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
ERROR: expected 4 dimensions, not 5
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
ERROR: Vector must have at least 3 dimensions
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
sparsevec
-----------------
{1:1.5,3:3.5}/5
sparsevec
---------------
{1:1.5,3:3.5}
(1 row)
SELECT '{0:0,1:1,2:0}/3'::sparsevec;
SELECT '{0:0,1:1,2:0}'::sparsevec;
sparsevec
-----------
{1:1}/3
{1:1}
(1 row)
SELECT '{1:1,0:1}/2'::sparsevec;
SELECT '{1:1,0:1}'::sparsevec;
ERROR: indexes must be in ascending order
LINE 1: SELECT '{1:1,0:1}/2'::sparsevec;
LINE 1: SELECT '{1:1,0:1}'::sparsevec;
^
SELECT '{}/5'::sparsevec;
SELECT '{}'::sparsevec;
sparsevec
-----------
{}/5
{}
(1 row)
SELECT '{}/-1'::sparsevec;
ERROR: sparsevec must have at least 1 dimension
LINE 1: SELECT '{}/-1'::sparsevec;
^
SELECT '{}/100001'::sparsevec;
ERROR: sparsevec cannot have more than 100000 dimensions
LINE 1: SELECT '{}/100001'::sparsevec;
^
SELECT '{}/16001'::sparsevec::vector;
ERROR: vector cannot have more than 16000 dimensions
SELECT '{-1:1}/1'::sparsevec;
SELECT '{}'::sparsevec::vector;
ERROR: vector must have at least 1 dimension
SELECT '{-1:1}'::sparsevec;
ERROR: index "-1" is out of range for type sparsevec
LINE 1: SELECT '{-1:1}/1'::sparsevec;
LINE 1: SELECT '{-1:1}'::sparsevec;
^
SELECT '{1:1}/1'::sparsevec;
ERROR: index must be less than dimensions
LINE 1: SELECT '{1:1}/1'::sparsevec;
^
SELECT '{}/1'::sparsevec(2);
ERROR: expected 2 dimensions, not 1
SELECT '{1:1}'::sparsevec;
sparsevec
-----------
{1:1}
(1 row)
SELECT '{}'::sparsevec(2);
ERROR: type modifier is not allowed for type "sparsevec"
LINE 1: SELECT '{}'::sparsevec(2);
^

10
test/sql/hnsw_sparsevec_cosine.sql
View File

@@ -1,13 +1,13 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
DROP TABLE t;

8
test/sql/hnsw_sparsevec_ip.sql
View File

@@ -1,12 +1,12 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
DROP TABLE t;

12
test/sql/hnsw_sparsevec_l2.sql
View File

@@ -1,22 +1,22 @@
SET enable_seqscan = off;
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
DROP TABLE t;
-- TODO move
CREATE TABLE t (val sparsevec(1001));
CREATE TABLE t (val sparsevec);
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
TRUNCATE t;

21
test/sql/sparsevec_functions.sql
View File

@@ -1,13 +1,12 @@
SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
SELECT l2_distance('{}'::sparsevec, '{1:1}');
SELECT '{}'::sparsevec <-> '{0:3,1:4}';
SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
SELECT cosine_distance('{}'::sparsevec, '{}');

25
test/sql/sparsevec_input.sql
View File

@@ -1,19 +1,18 @@
SELECT '{0:1.5,2:3.5}/5'::sparsevec;
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
SELECT '{0:1.5,2:3.5}'::sparsevec;
SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
SELECT '{0:0,1:1,2:0}/3'::sparsevec;
SELECT '{0:0,1:1,2:0}'::sparsevec;
SELECT '{1:1,0:1}/2'::sparsevec;
SELECT '{1:1,0:1}'::sparsevec;
SELECT '{}/5'::sparsevec;
SELECT '{}/-1'::sparsevec;
SELECT '{}/100001'::sparsevec;
SELECT '{}/16001'::sparsevec::vector;
SELECT '{}'::sparsevec;
SELECT '{}'::sparsevec::vector;
SELECT '{-1:1}/1'::sparsevec;
SELECT '{1:1}/1'::sparsevec;
SELECT '{-1:1}'::sparsevec;
SELECT '{1:1}'::sparsevec;
SELECT '{}/1'::sparsevec(2);
SELECT '{}'::sparsevec(2);