mirror of
https://github.com/pgvector/pgvector.git
synced 2026-07-22 20:15:46 +08:00
Compare commits
1 Commits
windows-si
...
concat-v2
| Author | SHA1 | Date | |
|---|---|---|---|
|
3f7c68f192 |
2
.github/workflows/build.yml
vendored
2
.github/workflows/build.yml
vendored
@@ -123,6 +123,6 @@ jobs:
|
||||
- uses: ankane/setup-postgres-valgrind@v1
|
||||
with:
|
||||
postgres-version: 16
|
||||
- run: make OPTFLAGS=""
|
||||
- run: make
|
||||
- run: sudo --preserve-env=PG_CONFIG make install
|
||||
- run: make installcheck
|
||||
|
||||
@@ -1,12 +1,11 @@
|
||||
## 0.7.0 (unreleased)
|
||||
|
||||
- Added `halfvec` type
|
||||
- Added `sparsevec` type
|
||||
- Added support for `bit` vectors to HNSW
|
||||
- Added support for bit vectors to HNSW
|
||||
- Added `hamming_distance` function
|
||||
- Added `jaccard_distance` function
|
||||
- Added `quantize_binary` function
|
||||
- Added `subvector` function
|
||||
- Added concatenate operator for vectors
|
||||
- Updated comparison operators to support vectors with different dimensions
|
||||
|
||||
## 0.6.2 (2024-03-18)
|
||||
|
||||
4
Makefile
4
Makefile
@@ -3,8 +3,8 @@ EXTVERSION = 0.6.2
|
||||
|
||||
MODULE_big = vector
|
||||
DATA = $(wildcard sql/*--*.sql)
|
||||
OBJS = src/bitvector.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
|
||||
HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
|
||||
OBJS = src/bitvector.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
|
||||
HEADERS = src/vector.h
|
||||
|
||||
TESTS = $(wildcard test/sql/*.sql)
|
||||
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
|
||||
|
||||
@@ -1,10 +1,10 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.6.2
|
||||
|
||||
OBJS = src\bitvector.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
|
||||
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
|
||||
OBJS = src\bitvector.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
|
||||
HEADERS = src\vector.h
|
||||
|
||||
REGRESS = bit_functions btree cast copy halfvec_functions halfvec_input hnsw_bit_hamming hnsw_bit_jaccard hnsw_halfvec_cosine hnsw_halfvec_ip hnsw_halfvec_l2 hnsw_options hnsw_sparsevec_cosine hnsw_sparsevec_ip hnsw_sparsevec_l2 hnsw_unlogged hnsw_vector_cosine hnsw_vector_ip hnsw_vector_l2 ivfflat_options ivfflat_unlogged ivfflat_vector_cosine ivfflat_vector_ip ivfflat_vector_l2 sparsevec_functions sparsevec_input vector_functions vector_input
|
||||
REGRESS = bit_functions btree cast copy hnsw_cosine hnsw_hamming hnsw_ip hnsw_jaccard hnsw_l2 hnsw_options hnsw_unlogged ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged vector_functions vector_input
|
||||
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
|
||||
|
||||
# For /arch flags
|
||||
|
||||
169
README.md
169
README.md
@@ -233,12 +233,7 @@ Jaccard distance - unreleased
|
||||
CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
|
||||
```
|
||||
|
||||
Supported types are:
|
||||
|
||||
- `vector` - up to 2,000 dimensions
|
||||
- `halfvec` - up to 4,000 dimensions (unreleased)
|
||||
- `bit` - up to 64,000 dimensions (unreleased)
|
||||
- `sparsevec` - up to 1,000 non-zero elements (unreleased)
|
||||
Vectors with up to 2,000 dimensions can be indexed, or bit vectors with up to 64,000 dimensions.
|
||||
|
||||
### Index Options
|
||||
|
||||
@@ -343,9 +338,7 @@ Cosine distance
|
||||
CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
|
||||
```
|
||||
|
||||
Supported types are:
|
||||
|
||||
- `vector` - up to 2,000 dimensions
|
||||
Vectors with up to 2,000 dimensions can be indexed.
|
||||
|
||||
### Query Options
|
||||
|
||||
@@ -419,103 +412,6 @@ Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html
|
||||
CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
|
||||
```
|
||||
|
||||
## Half Vectors
|
||||
|
||||
*Unreleased*
|
||||
|
||||
Use the `halfvec` type to store half-precision vectors
|
||||
|
||||
```sql
|
||||
CREATE TABLE items (id bigserial PRIMARY KEY, embedding halfvec(3));
|
||||
```
|
||||
|
||||
## Half Indexing
|
||||
|
||||
*Unreleased*
|
||||
|
||||
Index vectors at half precision for smaller indexes and faster build times
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw ((embedding::halfvec(3)) halfvec_l2_ops);
|
||||
```
|
||||
|
||||
Get the nearest neighbors
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY embedding::halfvec(3) <-> '[1,2,3]' LIMIT 5;
|
||||
```
|
||||
|
||||
## Binary Vectors
|
||||
|
||||
Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py))
|
||||
|
||||
```sql
|
||||
CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
|
||||
INSERT INTO items (embedding) VALUES ('000'), ('111');
|
||||
```
|
||||
|
||||
Get the nearest neighbors by Hamming distance
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY bit_count(embedding # '101') LIMIT 5;
|
||||
```
|
||||
|
||||
Or (unreleased)
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
|
||||
```
|
||||
|
||||
Also supports Jaccard distance (`<%>`)
|
||||
|
||||
## Binary Quantization
|
||||
|
||||
*Unreleased*
|
||||
|
||||
Use expression indexing for binary quantization
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw ((quantize_binary(embedding)::bit(3)) bit_hamming_ops);
|
||||
```
|
||||
|
||||
Get the nearest neighbors by Hamming distance
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY quantize_binary(embedding)::bit(3) <~> quantize_binary('[1,-2,3]') LIMIT 5;
|
||||
```
|
||||
|
||||
Re-rank by the original vectors for better recall
|
||||
|
||||
```sql
|
||||
SELECT * FROM (
|
||||
SELECT * FROM items ORDER BY quantize_binary(embedding)::bit(3) <~> quantize_binary('[1,-2,3]') LIMIT 20
|
||||
) ORDER BY embedding <=> '[1,-2,3]' LIMIT 5;
|
||||
```
|
||||
|
||||
## Sparse Vectors
|
||||
|
||||
*Unreleased*
|
||||
|
||||
Use the `sparsevec` type to store sparse vectors
|
||||
|
||||
```sql
|
||||
CREATE TABLE items (id bigserial PRIMARY KEY, embedding sparsevec(5));
|
||||
```
|
||||
|
||||
Insert vectors
|
||||
|
||||
```sql
|
||||
INSERT INTO items (embedding) VALUES ('{1:1,3:2,5:3}/5'), ('{1:4,3:5,5:6}/5');
|
||||
```
|
||||
|
||||
Note: The format is `{index1:value1,index2:value2,...}/dimensions` and indices start at 1 like SQL arrays
|
||||
|
||||
Get the nearest neighbors by L2 distance
|
||||
|
||||
```sql
|
||||
SELECT * FROM items ORDER BY embedding <-> '{1:3,3:1,5:2}/5' LIMIT 5;
|
||||
```
|
||||
|
||||
## Hybrid Search
|
||||
|
||||
Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search.
|
||||
@@ -732,6 +628,18 @@ and query with:
|
||||
SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
|
||||
```
|
||||
|
||||
#### Are binary vectors supported?
|
||||
|
||||
You can store binary vectors and perform exact nearest neighbor search by Hamming distance in Postgres without an extension ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py)).
|
||||
|
||||
```tsql
|
||||
CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
|
||||
INSERT INTO items (embedding) VALUES (B'000'), (B'111');
|
||||
SELECT * FROM items ORDER BY bit_count(embedding # B'101') LIMIT 5;
|
||||
```
|
||||
|
||||
Indexing is not currently supported.
|
||||
|
||||
#### Do indexes need to fit into memory?
|
||||
|
||||
No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
|
||||
@@ -804,9 +712,7 @@ Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosi
|
||||
## Reference
|
||||
|
||||
- [Vector](#vector-type)
|
||||
- [Halfvec](#halfvec-type)
|
||||
- [Bit](#bit-type)
|
||||
- [Sparsevec](#sparsevec-type)
|
||||
|
||||
### Vector Type
|
||||
|
||||
@@ -819,6 +725,7 @@ Operator | Description | Added
|
||||
\+ | element-wise addition |
|
||||
\- | element-wise subtraction |
|
||||
\* | element-wise multiplication | 0.5.0
|
||||
\|\| | concatenate | unreleased
|
||||
<-> | Euclidean distance |
|
||||
<#> | negative inner product |
|
||||
<=> | cosine distance |
|
||||
@@ -843,32 +750,9 @@ Function | Description | Added
|
||||
avg(vector) → vector | average |
|
||||
sum(vector) → vector | sum | 0.5.0
|
||||
|
||||
### Halfvec Type
|
||||
|
||||
Each half vector takes `2 * dimensions + 8` bytes of storage. Each element is a half-precision floating-point number, and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Half vectors can have up to 16,000 dimensions.
|
||||
|
||||
### Halfvec Operators
|
||||
|
||||
Operator | Description | Added
|
||||
--- | --- | ---
|
||||
<-> | Euclidean distance | unreleased
|
||||
<#> | negative inner product | unreleased
|
||||
<=> | cosine distance | unreleased
|
||||
|
||||
### Halfvec Functions
|
||||
|
||||
Function | Description | Added
|
||||
--- | --- | ---
|
||||
cosine_distance(halfvec, halfvec) → double precision | cosine distance | unreleased
|
||||
inner_product(halfvec, halfvec) → double precision | inner product | unreleased
|
||||
l2_distance(halfvec, halfvec) → double precision | Euclidean distance | unreleased
|
||||
l1_distance(halfvec, halfvec) → double precision | taxicab distance | unreleased
|
||||
quantize_binary(halfvec) → bit | quantize | unreleased
|
||||
subvector(halfvec, integer, integer) → halfvec | subvector | unreleased
|
||||
|
||||
### Bit Type
|
||||
|
||||
Each bit vector takes `dimensions / 8 + 8` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info.
|
||||
Each bit vector takes `dimensions / 8 + (5 or 8)` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info.
|
||||
|
||||
### Bit Operators
|
||||
|
||||
@@ -884,27 +768,6 @@ Function | Description | Added
|
||||
hamming_distance(bit, bit) → double precision | Hamming distance | unreleased
|
||||
jaccard_distance(bit, bit) → double precision | Jaccard distance | unreleased
|
||||
|
||||
### Sparsevec Type
|
||||
|
||||
Each sparse vector takes `8 * non-zero elements + 16` bytes of storage. Each element is a single-precision floating-point number, and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Sparse vectors can have up to 16,000 non-zero elements.
|
||||
|
||||
### Sparsevec Operators
|
||||
|
||||
Operator | Description | Added
|
||||
--- | --- | ---
|
||||
<-> | Euclidean distance | unreleased
|
||||
<#> | negative inner product | unreleased
|
||||
<=> | cosine distance | unreleased
|
||||
|
||||
### Sparsevec Functions
|
||||
|
||||
Function | Description | Added
|
||||
--- | --- | ---
|
||||
cosine_distance(sparsevec, sparsevec) → double precision | cosine distance | unreleased
|
||||
inner_product(sparsevec, sparsevec) → double precision | inner product | unreleased
|
||||
l2_distance(sparsevec, sparsevec) → double precision | Euclidean distance | unreleased
|
||||
l1_distance(sparsevec, sparsevec) → double precision | taxicab distance | unreleased
|
||||
|
||||
## Installation Notes - Linux and Mac
|
||||
|
||||
### Postgres Location
|
||||
|
||||
@@ -7,6 +7,13 @@ CREATE FUNCTION quantize_binary(vector) RETURNS bit
|
||||
CREATE FUNCTION subvector(vector, int, int) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE OPERATOR || (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
|
||||
);
|
||||
|
||||
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
@@ -32,228 +39,3 @@ CREATE OPERATOR CLASS bit_jaccard_ops
|
||||
FOR TYPE bit USING hnsw AS
|
||||
OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 jaccard_distance(bit, bit);
|
||||
|
||||
CREATE TYPE halfvec;
|
||||
|
||||
CREATE FUNCTION halfvec_in(cstring, oid, integer) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_out(halfvec) RETURNS cstring
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_typmod_in(cstring[]) RETURNS integer
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_recv(internal, oid, integer) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_send(halfvec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE halfvec (
|
||||
INPUT = halfvec_in,
|
||||
OUTPUT = halfvec_out,
|
||||
TYPMOD_IN = halfvec_typmod_in,
|
||||
RECEIVE = halfvec_recv,
|
||||
SEND = halfvec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
CREATE FUNCTION l2_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_norm(halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION quantize_binary(halfvec) RETURNS bit
|
||||
AS 'MODULE_PATHNAME', 'halfvec_quantize_binary' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION subvector(halfvec, int, int) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME', 'halfvec_subvector' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_l2_squared_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_negative_inner_product(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec(halfvec, integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(integer[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(real[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(double precision[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(numeric[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_to_float4(halfvec, integer, boolean) RETURNS real[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE CAST (halfvec AS halfvec)
|
||||
WITH FUNCTION halfvec(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (halfvec AS real[])
|
||||
WITH FUNCTION halfvec_to_float4(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (integer[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(integer[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (real[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(real[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (double precision[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(double precision[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (numeric[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(numeric[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_l2_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec);
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_ip_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec);
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_cosine_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
|
||||
FUNCTION 2 halfvec_norm(halfvec);
|
||||
|
||||
CREATE FUNCTION halfvec_to_vector(halfvec, integer, boolean) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_to_halfvec(vector, integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE CAST (halfvec AS vector)
|
||||
WITH FUNCTION halfvec_to_vector(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (vector AS halfvec)
|
||||
WITH FUNCTION vector_to_halfvec(vector, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE TYPE sparsevec;
|
||||
|
||||
CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
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;
|
||||
|
||||
CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE sparsevec (
|
||||
INPUT = sparsevec_in,
|
||||
OUTPUT = sparsevec_out,
|
||||
TYPMOD_IN = sparsevec_typmod_in,
|
||||
RECEIVE = sparsevec_recv,
|
||||
SEND = sparsevec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
CREATE FUNCTION l2_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_l2_squared_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_negative_inner_product(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec(sparsevec, integer, boolean) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_to_sparsevec(vector, integer, boolean) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE CAST (sparsevec AS sparsevec)
|
||||
WITH FUNCTION sparsevec(sparsevec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (sparsevec AS vector)
|
||||
WITH FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (vector AS sparsevec)
|
||||
WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_l2_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <-> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_l2_squared_distance(sparsevec, sparsevec);
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_ip_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <#> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec);
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_cosine_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
|
||||
FUNCTION 2 sparsevec_norm(sparsevec);
|
||||
|
||||
262
sql/vector.sql
262
sql/vector.sql
@@ -105,6 +105,9 @@ CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
|
||||
CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- vector aggregates
|
||||
|
||||
CREATE AGGREGATE avg(vector) (
|
||||
@@ -194,6 +197,10 @@ CREATE OPERATOR * (
|
||||
COMMUTATOR = *
|
||||
);
|
||||
|
||||
CREATE OPERATOR || (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
|
||||
);
|
||||
|
||||
CREATE OPERATOR < (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_lt,
|
||||
COMMUTATOR = > , NEGATOR = >= ,
|
||||
@@ -321,258 +328,3 @@ CREATE OPERATOR CLASS bit_jaccard_ops
|
||||
FOR TYPE bit USING hnsw AS
|
||||
OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 jaccard_distance(bit, bit);
|
||||
|
||||
-- halfvec type
|
||||
|
||||
CREATE TYPE halfvec;
|
||||
|
||||
CREATE FUNCTION halfvec_in(cstring, oid, integer) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_out(halfvec) RETURNS cstring
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_typmod_in(cstring[]) RETURNS integer
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_recv(internal, oid, integer) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_send(halfvec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE halfvec (
|
||||
INPUT = halfvec_in,
|
||||
OUTPUT = halfvec_out,
|
||||
TYPMOD_IN = halfvec_typmod_in,
|
||||
RECEIVE = halfvec_recv,
|
||||
SEND = halfvec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
-- halfvec functions
|
||||
|
||||
CREATE FUNCTION l2_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_norm(halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION quantize_binary(halfvec) RETURNS bit
|
||||
AS 'MODULE_PATHNAME', 'halfvec_quantize_binary' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION subvector(halfvec, int, int) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME', 'halfvec_subvector' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- halfvec private functions
|
||||
|
||||
CREATE FUNCTION halfvec_l2_squared_distance(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_negative_inner_product(halfvec, halfvec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- halfvec cast functions
|
||||
|
||||
CREATE FUNCTION halfvec(halfvec, integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(integer[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(real[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(double precision[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION array_to_halfvec(numeric[], integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION halfvec_to_float4(halfvec, integer, boolean) RETURNS real[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- halfvec casts
|
||||
|
||||
CREATE CAST (halfvec AS halfvec)
|
||||
WITH FUNCTION halfvec(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (halfvec AS real[])
|
||||
WITH FUNCTION halfvec_to_float4(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (integer[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(integer[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (real[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(real[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (double precision[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(double precision[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
CREATE CAST (numeric[] AS halfvec)
|
||||
WITH FUNCTION array_to_halfvec(numeric[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
-- halfvec operators
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
-- halfvec opclasses
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_l2_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec);
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_ip_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec);
|
||||
|
||||
CREATE OPERATOR CLASS halfvec_cosine_ops
|
||||
FOR TYPE halfvec USING hnsw AS
|
||||
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
|
||||
FUNCTION 2 halfvec_norm(halfvec);
|
||||
|
||||
-- extension casts
|
||||
|
||||
CREATE FUNCTION halfvec_to_vector(halfvec, integer, boolean) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_to_halfvec(vector, integer, boolean) RETURNS halfvec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE CAST (halfvec AS vector)
|
||||
WITH FUNCTION halfvec_to_vector(halfvec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (vector AS halfvec)
|
||||
WITH FUNCTION vector_to_halfvec(vector, integer, boolean) AS IMPLICIT;
|
||||
|
||||
--- sparsevec type
|
||||
|
||||
CREATE TYPE sparsevec;
|
||||
|
||||
CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
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;
|
||||
|
||||
CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE TYPE sparsevec (
|
||||
INPUT = sparsevec_in,
|
||||
OUTPUT = sparsevec_out,
|
||||
TYPMOD_IN = sparsevec_typmod_in,
|
||||
RECEIVE = sparsevec_recv,
|
||||
SEND = sparsevec_send,
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
-- sparsevec functions
|
||||
|
||||
CREATE FUNCTION l2_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- sparsevec private functions
|
||||
|
||||
CREATE FUNCTION sparsevec_l2_squared_distance(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_negative_inner_product(sparsevec, sparsevec) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- sparsevec cast functions
|
||||
|
||||
CREATE FUNCTION sparsevec(sparsevec, integer, boolean) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION vector_to_sparsevec(vector, integer, boolean) RETURNS sparsevec
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- sparsevec casts
|
||||
|
||||
CREATE CAST (sparsevec AS sparsevec)
|
||||
WITH FUNCTION sparsevec(sparsevec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (sparsevec AS vector)
|
||||
WITH FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) AS IMPLICIT;
|
||||
|
||||
CREATE CAST (vector AS sparsevec)
|
||||
WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
|
||||
|
||||
-- sparsevec operators
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <#> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_negative_inner_product,
|
||||
COMMUTATOR = '<#>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <=> (
|
||||
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = cosine_distance,
|
||||
COMMUTATOR = '<=>'
|
||||
);
|
||||
|
||||
-- sparsevec opclasses
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_l2_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <-> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_l2_squared_distance(sparsevec, sparsevec);
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_ip_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <#> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec);
|
||||
|
||||
CREATE OPERATOR CLASS sparsevec_cosine_ops
|
||||
FOR TYPE sparsevec USING hnsw AS
|
||||
OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
|
||||
FUNCTION 2 sparsevec_norm(sparsevec);
|
||||
|
||||
1095
src/halfvec.c
1095
src/halfvec.c
File diff suppressed because it is too large
Load Diff
@@ -1,45 +0,0 @@
|
||||
#ifndef HALFVEC_H
|
||||
#define HALFVEC_H
|
||||
|
||||
#define __STDC_WANT_IEC_60559_TYPES_EXT__
|
||||
|
||||
#include <float.h>
|
||||
|
||||
#include "vector.h"
|
||||
|
||||
/* F16C has better performance than _Float16 (on x86-64) */
|
||||
#if defined(__F16C__) || defined(_MSC_VER)
|
||||
#define F16C_SUPPORT
|
||||
#elif defined(__FLT16_MAX__)
|
||||
#define FLT16_SUPPORT
|
||||
#endif
|
||||
|
||||
#ifdef FLT16_SUPPORT
|
||||
#define half _Float16
|
||||
#define HALF_MAX FLT16_MAX
|
||||
#else
|
||||
#define half uint16
|
||||
#define HALF_MAX 65504
|
||||
#endif
|
||||
|
||||
#define HALFVEC_MAX_DIM VECTOR_MAX_DIM
|
||||
|
||||
#define HALFVEC_SIZE(_dim) (offsetof(HalfVector, x) + sizeof(half)*(_dim))
|
||||
#define DatumGetHalfVector(x) ((HalfVector *) PG_DETOAST_DATUM(x))
|
||||
#define PG_GETARG_HALFVEC_P(x) DatumGetHalfVector(PG_GETARG_DATUM(x))
|
||||
#define PG_RETURN_HALFVEC_P(x) PG_RETURN_POINTER(x)
|
||||
|
||||
typedef struct HalfVector
|
||||
{
|
||||
int32 vl_len_; /* varlena header (do not touch directly!) */
|
||||
int16 dim; /* number of dimensions */
|
||||
int16 unused;
|
||||
half x[FLEXIBLE_ARRAY_MEMBER];
|
||||
} HalfVector;
|
||||
|
||||
HalfVector *InitHalfVector(int dim);
|
||||
float HalfToFloat4(half num);
|
||||
half Float4ToHalf(float num);
|
||||
half Float4ToHalfUnchecked(float num);
|
||||
|
||||
#endif
|
||||
@@ -17,7 +17,6 @@
|
||||
#endif
|
||||
|
||||
#define HNSW_MAX_DIM 2000
|
||||
#define HNSW_MAX_NNZ 1000
|
||||
|
||||
/* Support functions */
|
||||
#define HNSW_DISTANCE_PROC 1
|
||||
@@ -59,9 +58,7 @@
|
||||
typedef enum HnswType
|
||||
{
|
||||
HNSW_TYPE_VECTOR,
|
||||
HNSW_TYPE_HALFVEC,
|
||||
HNSW_TYPE_BIT,
|
||||
HNSW_TYPE_SPARSEVEC
|
||||
HNSW_TYPE_BIT
|
||||
} HnswType;
|
||||
|
||||
/* Build phases */
|
||||
@@ -378,7 +375,6 @@ int HnswGetEfConstruction(Relation index);
|
||||
FmgrInfo *HnswOptionalProcInfo(Relation index, uint16 procnum);
|
||||
HnswType HnswGetType(Relation index);
|
||||
bool HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type);
|
||||
void HnswCheckValue(Datum value, HnswType type);
|
||||
Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
|
||||
void HnswInitPage(Buffer buf, Page page);
|
||||
void HnswInit(void);
|
||||
|
||||
@@ -487,9 +487,6 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
|
||||
/* Detoast once for all calls */
|
||||
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
|
||||
|
||||
/* Check value */
|
||||
HnswCheckValue(value, buildstate->type);
|
||||
|
||||
/* Normalize if needed */
|
||||
if (buildstate->normprocinfo != NULL)
|
||||
{
|
||||
@@ -677,12 +674,8 @@ GetMaxDimensions(HnswType type)
|
||||
{
|
||||
int maxDimensions = HNSW_MAX_DIM;
|
||||
|
||||
if (type == HNSW_TYPE_HALFVEC)
|
||||
maxDimensions *= 2;
|
||||
else if (type == HNSW_TYPE_BIT)
|
||||
if (type == HNSW_TYPE_BIT)
|
||||
maxDimensions *= 32;
|
||||
else if (type == HNSW_TYPE_SPARSEVEC)
|
||||
maxDimensions = INT_MAX;
|
||||
|
||||
return maxDimensions;
|
||||
}
|
||||
|
||||
@@ -614,19 +614,15 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
|
||||
Datum value;
|
||||
FmgrInfo *normprocinfo;
|
||||
Oid collation = index->rd_indcollation[0];
|
||||
HnswType type = HnswGetType(index);
|
||||
|
||||
/* Detoast once for all calls */
|
||||
value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
|
||||
|
||||
/* Check value */
|
||||
HnswCheckValue(value, type);
|
||||
|
||||
/* Normalize if needed */
|
||||
normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
|
||||
if (normprocinfo != NULL)
|
||||
{
|
||||
if (!HnswNormValue(normprocinfo, collation, &value, type))
|
||||
if (!HnswNormValue(normprocinfo, collation, &value, HnswGetType(index)))
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
#include "postgres.h"
|
||||
|
||||
#include "access/relscan.h"
|
||||
#include "bitvector.h"
|
||||
#include "catalog/pg_type_d.h"
|
||||
#include "hnsw.h"
|
||||
#include "pgstat.h"
|
||||
#include "storage/bufmgr.h"
|
||||
@@ -158,10 +160,6 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
|
||||
|
||||
so->first = false;
|
||||
|
||||
#if defined(HNSW_MEMORY) && PG_VERSION_NUM >= 130000
|
||||
elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(so->tmpCtx, false) / (1024 * 1024));
|
||||
#endif
|
||||
}
|
||||
|
||||
while (list_length(so->w) > 0)
|
||||
|
||||
@@ -3,17 +3,13 @@
|
||||
#include <math.h>
|
||||
|
||||
#include "access/generic_xlog.h"
|
||||
#include "catalog/pg_type.h"
|
||||
#include "catalog/pg_type_d.h"
|
||||
#include "halfvec.h"
|
||||
#include "hnsw.h"
|
||||
#include "lib/pairingheap.h"
|
||||
#include "sparsevec.h"
|
||||
#include "storage/bufmgr.h"
|
||||
#include "utils/datum.h"
|
||||
#include "utils/memdebug.h"
|
||||
#include "utils/rel.h"
|
||||
#include "utils/syscache.h"
|
||||
#include "vector.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 130000
|
||||
@@ -161,33 +157,11 @@ HnswType
|
||||
HnswGetType(Relation index)
|
||||
{
|
||||
Oid typid = TupleDescAttr(index->rd_att, 0)->atttypid;
|
||||
HeapTuple tuple;
|
||||
Form_pg_type type;
|
||||
HnswType result;
|
||||
|
||||
if (typid == BITOID)
|
||||
if (typid == BITOID || typid == VARBITOID)
|
||||
return HNSW_TYPE_BIT;
|
||||
|
||||
tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(typid));
|
||||
if (!HeapTupleIsValid(tuple))
|
||||
elog(ERROR, "cache lookup failed for type %u", typid);
|
||||
|
||||
type = (Form_pg_type) GETSTRUCT(tuple);
|
||||
if (strcmp(NameStr(type->typname), "vector") == 0)
|
||||
result = HNSW_TYPE_VECTOR;
|
||||
else if (strcmp(NameStr(type->typname), "halfvec") == 0)
|
||||
result = HNSW_TYPE_HALFVEC;
|
||||
else if (strcmp(NameStr(type->typname), "sparsevec") == 0)
|
||||
result = HNSW_TYPE_SPARSEVEC;
|
||||
else
|
||||
{
|
||||
ReleaseSysCache(tuple);
|
||||
elog(ERROR, "type not supported for hnsw index");
|
||||
}
|
||||
|
||||
ReleaseSysCache(tuple);
|
||||
|
||||
return result;
|
||||
return HNSW_TYPE_VECTOR;
|
||||
}
|
||||
|
||||
/*
|
||||
@@ -216,34 +190,6 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
}
|
||||
else if (type == HNSW_TYPE_HALFVEC)
|
||||
{
|
||||
HalfVector *v = DatumGetHalfVector(*value);
|
||||
HalfVector *result = InitHalfVector(v->dim);
|
||||
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
{
|
||||
/* TODO Fix */
|
||||
result->x[i] = Float4ToHalfUnchecked(HalfToFloat4(v->x[i]) / norm);
|
||||
}
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
}
|
||||
else if (type == HNSW_TYPE_SPARSEVEC)
|
||||
{
|
||||
SparseVector *v = DatumGetSparseVector(*value);
|
||||
SparseVector *result = InitSparseVector(v->dim, v->nnz);
|
||||
float *vx = SPARSEVEC_VALUES(v);
|
||||
float *rx = SPARSEVEC_VALUES(result);
|
||||
|
||||
for (int i = 0; i < v->nnz; i++)
|
||||
{
|
||||
result->indices[i] = v->indices[i];
|
||||
rx[i] = vx[i] / norm;
|
||||
}
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
}
|
||||
else
|
||||
elog(ERROR, "Unsupported type");
|
||||
|
||||
@@ -253,21 +199,6 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
|
||||
return false;
|
||||
}
|
||||
|
||||
/*
|
||||
* Check if a value can be indexed
|
||||
*/
|
||||
void
|
||||
HnswCheckValue(Datum value, HnswType type)
|
||||
{
|
||||
if (type == HNSW_TYPE_SPARSEVEC)
|
||||
{
|
||||
SparseVector *vec = DatumGetSparseVector(value);
|
||||
|
||||
if (vec->nnz > HNSW_MAX_NNZ)
|
||||
elog(ERROR, "sparsevec cannot have more than %d non-zero elements for hnsw index", HNSW_MAX_NNZ);
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* New buffer
|
||||
*/
|
||||
|
||||
@@ -57,7 +57,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
|
||||
*/
|
||||
if (buildstate->kmeansnormprocinfo != NULL)
|
||||
{
|
||||
if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->type))
|
||||
if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value))
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -153,7 +153,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
|
||||
/* Normalize if needed */
|
||||
if (buildstate->normprocinfo != NULL)
|
||||
{
|
||||
if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
|
||||
if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value))
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -312,39 +312,25 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Get max dimensions
|
||||
*/
|
||||
static int
|
||||
GetMaxDimensions(IvfflatType type)
|
||||
{
|
||||
return IVFFLAT_MAX_DIM;
|
||||
}
|
||||
|
||||
/*
|
||||
* Initialize the build state
|
||||
*/
|
||||
static void
|
||||
InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo)
|
||||
{
|
||||
int maxDimensions;
|
||||
|
||||
buildstate->heap = heap;
|
||||
buildstate->index = index;
|
||||
buildstate->indexInfo = indexInfo;
|
||||
buildstate->type = IvfflatGetType(index);
|
||||
|
||||
buildstate->lists = IvfflatGetLists(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->dimensions > maxDimensions)
|
||||
elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", maxDimensions);
|
||||
if (buildstate->dimensions > IVFFLAT_MAX_DIM)
|
||||
elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", IVFFLAT_MAX_DIM);
|
||||
|
||||
buildstate->reltuples = 0;
|
||||
buildstate->indtuples = 0;
|
||||
|
||||
@@ -43,11 +43,6 @@
|
||||
#define IVFFLAT_MAX_LISTS 32768
|
||||
#define IVFFLAT_DEFAULT_PROBES 1
|
||||
|
||||
typedef enum IvfflatType
|
||||
{
|
||||
IVFFLAT_TYPE_VECTOR
|
||||
} IvfflatType;
|
||||
|
||||
/* Build phases */
|
||||
/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
|
||||
#define PROGRESS_IVFFLAT_PHASE_KMEANS 2
|
||||
@@ -158,7 +153,6 @@ typedef struct IvfflatBuildState
|
||||
Relation heap;
|
||||
Relation index;
|
||||
IndexInfo *indexInfo;
|
||||
IvfflatType type;
|
||||
|
||||
/* Settings */
|
||||
int dimensions;
|
||||
@@ -272,8 +266,7 @@ void VectorArrayFree(VectorArray arr);
|
||||
void PrintVectorArray(char *msg, VectorArray arr);
|
||||
void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
|
||||
FmgrInfo *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
|
||||
IvfflatType IvfflatGetType(Relation index);
|
||||
bool IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, IvfflatType type);
|
||||
bool IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value);
|
||||
int IvfflatGetLists(Relation index);
|
||||
void IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);
|
||||
void IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);
|
||||
|
||||
@@ -85,7 +85,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
|
||||
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
|
||||
if (normprocinfo != NULL)
|
||||
{
|
||||
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, IvfflatGetType(index)))
|
||||
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value))
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
@@ -268,7 +268,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
if (so->first)
|
||||
{
|
||||
Datum value;
|
||||
IvfflatType type = IvfflatGetType(scan->indexRelation);
|
||||
|
||||
/* Count index scan for stats */
|
||||
pgstat_count_index_scan(scan->indexRelation);
|
||||
@@ -283,12 +282,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
elog(ERROR, "non-MVCC snapshots are not supported with ivfflat");
|
||||
|
||||
if (scan->orderByData->sk_flags & SK_ISNULL)
|
||||
{
|
||||
if (type == IVFFLAT_TYPE_VECTOR)
|
||||
value = PointerGetDatum(InitVector(so->dimensions));
|
||||
else
|
||||
elog(ERROR, "Unsupported type");
|
||||
}
|
||||
value = PointerGetDatum(InitVector(so->dimensions));
|
||||
else
|
||||
{
|
||||
value = scan->orderByData->sk_argument;
|
||||
@@ -299,7 +293,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
|
||||
|
||||
/* Fine if normalization fails */
|
||||
if (so->normprocinfo != NULL)
|
||||
IvfflatNormValue(so->normprocinfo, so->collation, &value, type);
|
||||
IvfflatNormValue(so->normprocinfo, so->collation, &value);
|
||||
}
|
||||
|
||||
IvfflatBench("GetScanLists", GetScanLists(scan, value));
|
||||
|
||||
@@ -66,15 +66,6 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
|
||||
return index_getprocinfo(index, 1, procnum);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get type
|
||||
*/
|
||||
IvfflatType
|
||||
IvfflatGetType(Relation index)
|
||||
{
|
||||
return IVFFLAT_TYPE_VECTOR;
|
||||
}
|
||||
|
||||
/*
|
||||
* Divide by the norm
|
||||
*
|
||||
@@ -84,24 +75,19 @@ IvfflatGetType(Relation index)
|
||||
* if it's different than the original value
|
||||
*/
|
||||
bool
|
||||
IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, IvfflatType type)
|
||||
IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value)
|
||||
{
|
||||
double norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
|
||||
|
||||
if (norm > 0)
|
||||
{
|
||||
if (type == IVFFLAT_TYPE_VECTOR)
|
||||
{
|
||||
Vector *v = DatumGetVector(*value);
|
||||
Vector *result = InitVector(v->dim);
|
||||
Vector *v = DatumGetVector(*value);
|
||||
Vector *result = InitVector(v->dim);
|
||||
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
}
|
||||
else
|
||||
elog(ERROR, "Unsupported type");
|
||||
*value = PointerGetDatum(result);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
788
src/sparsevec.c
788
src/sparsevec.c
@@ -1,788 +0,0 @@
|
||||
#include "postgres.h"
|
||||
|
||||
#include <limits.h>
|
||||
#include <math.h>
|
||||
|
||||
#include "fmgr.h"
|
||||
#include "libpq/pqformat.h"
|
||||
#include "sparsevec.h"
|
||||
#include "utils/array.h"
|
||||
#include "utils/builtins.h"
|
||||
#include "vector.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 120000
|
||||
#include "common/shortest_dec.h"
|
||||
#include "utils/float.h"
|
||||
#else
|
||||
#include <float.h>
|
||||
#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)
|
||||
{
|
||||
if (nnz < 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("sparsevec cannot have negative number of elements")));
|
||||
|
||||
if (nnz > SPARSEVEC_MAX_NNZ)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
errmsg("sparsevec cannot have more than %d non-zero elements", SPARSEVEC_MAX_NNZ)));
|
||||
|
||||
if (nnz > dim)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
errmsg("sparsevec cannot have more elements than dimensions")));
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure valid index
|
||||
*/
|
||||
static inline void
|
||||
CheckIndex(int32 *indices, int i, int dim)
|
||||
{
|
||||
int32 index = indices[i];
|
||||
|
||||
if (index < 1)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("index must be greater than zero")));
|
||||
|
||||
if (index > dim)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("index must be less than or equal to dimensions")));
|
||||
|
||||
if (i > 0)
|
||||
{
|
||||
if (index < indices[i - 1])
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("indexes must be in ascending order")));
|
||||
|
||||
if (index == indices[i - 1])
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("indexes must not contain duplicates")));
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure finite element
|
||||
*/
|
||||
static inline void
|
||||
CheckElement(float value)
|
||||
{
|
||||
if (isnan(value))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("NaN not allowed in sparsevec")));
|
||||
|
||||
if (isinf(value))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("infinite value not allowed in sparsevec")));
|
||||
}
|
||||
|
||||
/*
|
||||
* Allocate and initialize a new sparse vector
|
||||
*/
|
||||
SparseVector *
|
||||
InitSparseVector(int dim, int nnz)
|
||||
{
|
||||
SparseVector *result;
|
||||
int size;
|
||||
|
||||
size = SPARSEVEC_SIZE(nnz);
|
||||
result = (SparseVector *) palloc0(size);
|
||||
SET_VARSIZE(result, size);
|
||||
result->dim = dim;
|
||||
result->nnz = nnz;
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/*
|
||||
* Check for whitespace, since array_isspace() is static
|
||||
*/
|
||||
static inline bool
|
||||
sparsevec_isspace(char ch)
|
||||
{
|
||||
if (ch == ' ' ||
|
||||
ch == '\t' ||
|
||||
ch == '\n' ||
|
||||
ch == '\r' ||
|
||||
ch == '\v' ||
|
||||
ch == '\f')
|
||||
return true;
|
||||
return false;
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert textual representation to internal representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_in);
|
||||
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;
|
||||
float *rvalues;
|
||||
char *litcopy = pstrdup(lit);
|
||||
char *str = litcopy;
|
||||
int32 *indices;
|
||||
float *values;
|
||||
int maxNnz;
|
||||
int nnz = 0;
|
||||
|
||||
maxNnz = 1;
|
||||
pt = str;
|
||||
while (*pt != '\0')
|
||||
{
|
||||
if (*pt == ',')
|
||||
maxNnz++;
|
||||
|
||||
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));
|
||||
|
||||
while (sparsevec_isspace(*str))
|
||||
str++;
|
||||
|
||||
if (*str != '{')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("malformed sparsevec literal: \"%s\"", lit),
|
||||
errdetail("Vector contents must start with \"{\".")));
|
||||
|
||||
str++;
|
||||
pt = strtok(str, ",");
|
||||
stringEnd = pt;
|
||||
|
||||
while (pt != NULL && *stringEnd != '}')
|
||||
{
|
||||
long index;
|
||||
float value;
|
||||
|
||||
/* TODO Better error */
|
||||
if (nnz == maxNnz)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("ran out of buffer: \"%s\"", lit)));
|
||||
|
||||
while (sparsevec_isspace(*pt))
|
||||
pt++;
|
||||
|
||||
/* Check for empty string like float4in */
|
||||
if (*pt == '\0')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
/* Use similar logic as int2vectorin */
|
||||
errno = 0;
|
||||
index = strtol(pt, &stringEnd, 10);
|
||||
|
||||
if (stringEnd == pt)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
if (errno == ERANGE || index < 1 || index > INT_MAX)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
|
||||
errmsg("index \"%ld\" is out of range for type sparsevec", index)));
|
||||
|
||||
if (stringEnd == pt)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
while (sparsevec_isspace(*stringEnd))
|
||||
stringEnd++;
|
||||
|
||||
if (*stringEnd != ':')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
stringEnd++;
|
||||
|
||||
while (sparsevec_isspace(*stringEnd))
|
||||
stringEnd++;
|
||||
|
||||
errno = 0;
|
||||
pt = stringEnd;
|
||||
value = strtof(pt, &stringEnd);
|
||||
|
||||
if (stringEnd == pt)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
/* Check for range error like float4in */
|
||||
if (errno == ERANGE && (value == 0 || isinf(value)))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
|
||||
errmsg("\"%s\" is out of range for type sparsevec", pt)));
|
||||
|
||||
/* TODO Decide whether to store zero values */
|
||||
if (value != 0)
|
||||
{
|
||||
indices[nnz] = index;
|
||||
values[nnz] = value;
|
||||
nnz++;
|
||||
}
|
||||
|
||||
if (*stringEnd != '\0' && *stringEnd != '}')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
|
||||
|
||||
pt = strtok(NULL, ",");
|
||||
}
|
||||
|
||||
if (stringEnd == NULL || *stringEnd != '}')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("malformed sparsevec literal: \"%s\"", lit),
|
||||
errdetail("Unexpected end of input.")));
|
||||
|
||||
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++;
|
||||
|
||||
if (*stringEnd != '\0')
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("malformed sparsevec literal: \"%s\"", lit),
|
||||
errdetail("Junk after closing.")));
|
||||
|
||||
pfree(litcopy);
|
||||
|
||||
CheckDim(dim);
|
||||
CheckExpectedDim(typmod, dim);
|
||||
|
||||
result = InitSparseVector(dim, 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);
|
||||
CheckElement(rvalues[i]);
|
||||
}
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
#define AppendChar(ptr, c) (*(ptr)++ = (c))
|
||||
#define AppendFloat(ptr, f) ((ptr) += float_to_shortest_decimal_bufn((f), (ptr)))
|
||||
|
||||
#if PG_VERSION_NUM >= 140000
|
||||
#define AppendInt(ptr, i) ((ptr) += pg_ltoa((i), (ptr)))
|
||||
#else
|
||||
#define AppendInt(ptr, i) \
|
||||
do { \
|
||||
pg_ltoa(i, ptr); \
|
||||
while (*ptr != '\0') \
|
||||
ptr++; \
|
||||
} while (0)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Convert internal representation to textual representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_out);
|
||||
Datum
|
||||
sparsevec_out(PG_FUNCTION_ARGS)
|
||||
{
|
||||
SparseVector *sparsevec = PG_GETARG_SPARSEVEC_P(0);
|
||||
float *values = SPARSEVEC_VALUES(sparsevec);
|
||||
char *buf;
|
||||
char *ptr;
|
||||
|
||||
/*
|
||||
* Need:
|
||||
*
|
||||
* nnz * 10 bytes for index (positive integer)
|
||||
*
|
||||
* nnz bytes for :
|
||||
*
|
||||
* nnz * (FLOAT_SHORTEST_DECIMAL_LEN - 1) bytes for
|
||||
* float_to_shortest_decimal_bufn
|
||||
*
|
||||
* nnz - 1 bytes for ,
|
||||
*
|
||||
* 10 bytes for dimensions
|
||||
*
|
||||
* 4 bytes for {, }, /, and \0
|
||||
*/
|
||||
buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 13);
|
||||
ptr = buf;
|
||||
|
||||
AppendChar(ptr, '{');
|
||||
|
||||
for (int i = 0; i < sparsevec->nnz; i++)
|
||||
{
|
||||
if (i > 0)
|
||||
AppendChar(ptr, ',');
|
||||
|
||||
AppendInt(ptr, sparsevec->indices[i]);
|
||||
AppendChar(ptr, ':');
|
||||
AppendFloat(ptr, values[i]);
|
||||
}
|
||||
|
||||
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
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_recv);
|
||||
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;
|
||||
float *values;
|
||||
|
||||
dim = pq_getmsgint(buf, sizeof(int32));
|
||||
nnz = pq_getmsgint(buf, sizeof(int32));
|
||||
unused = pq_getmsgint(buf, sizeof(int32));
|
||||
|
||||
CheckDim(dim);
|
||||
CheckNnz(nnz, dim);
|
||||
CheckExpectedDim(typmod, dim);
|
||||
|
||||
if (unused != 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("expected unused to be 0, not %d", unused)));
|
||||
|
||||
result = InitSparseVector(dim, 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);
|
||||
}
|
||||
|
||||
for (int i = 0; i < nnz; i++)
|
||||
{
|
||||
values[i] = pq_getmsgfloat4(buf);
|
||||
CheckElement(values[i]);
|
||||
}
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert internal representation to the external binary representation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_send);
|
||||
Datum
|
||||
sparsevec_send(PG_FUNCTION_ARGS)
|
||||
{
|
||||
SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
|
||||
float *values = SPARSEVEC_VALUES(svec);
|
||||
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));
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
pq_sendint(&buf, svec->indices[i], sizeof(int32));
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
pq_sendfloat4(&buf, values[i]);
|
||||
|
||||
PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
|
||||
}
|
||||
|
||||
/*
|
||||
* 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);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert dense vector to sparse vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_sparsevec);
|
||||
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);
|
||||
values = SPARSEVEC_VALUES(result);
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
if (vec->x[i] != 0)
|
||||
{
|
||||
/* Safety check */
|
||||
if (j == nnz)
|
||||
elog(ERROR, "safety check failed");
|
||||
|
||||
result->indices[j] = i + 1;
|
||||
values[j] = vec->x[i];
|
||||
j++;
|
||||
}
|
||||
}
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 squared distance between sparse vectors
|
||||
*/
|
||||
static double
|
||||
l2_distance_squared_internal(SparseVector * a, SparseVector * b)
|
||||
{
|
||||
float *ax = SPARSEVEC_VALUES(a);
|
||||
float *bx = SPARSEVEC_VALUES(b);
|
||||
double distance = 0.0;
|
||||
int bpos = 0;
|
||||
|
||||
for (int i = 0; i < a->nnz; i++)
|
||||
{
|
||||
int ai = a->indices[i];
|
||||
int bi = -1;
|
||||
|
||||
for (int j = bpos; j < b->nnz; j++)
|
||||
{
|
||||
bi = b->indices[j];
|
||||
|
||||
if (ai == bi)
|
||||
{
|
||||
double diff = ax[i] - bx[j];
|
||||
|
||||
distance += diff * diff;
|
||||
}
|
||||
else if (ai > bi)
|
||||
distance += bx[j] * bx[j];
|
||||
|
||||
/* Update start for next iteration */
|
||||
if (ai >= bi)
|
||||
bpos = j + 1;
|
||||
|
||||
/* Found or passed it */
|
||||
if (bi >= ai)
|
||||
break;
|
||||
}
|
||||
|
||||
if (ai != bi)
|
||||
distance += ax[i] * ax[i];
|
||||
}
|
||||
|
||||
for (int j = bpos; j < b->nnz; j++)
|
||||
distance += bx[j] * bx[j];
|
||||
|
||||
return distance;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 distance between sparse vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_distance);
|
||||
Datum
|
||||
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)));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 squared distance between sparse vectors
|
||||
* This saves a sqrt calculation
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_squared_distance);
|
||||
Datum
|
||||
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));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the inner product of two sparse vectors
|
||||
*/
|
||||
static double
|
||||
inner_product_internal(SparseVector * a, SparseVector * b)
|
||||
{
|
||||
float *ax = SPARSEVEC_VALUES(a);
|
||||
float *bx = SPARSEVEC_VALUES(b);
|
||||
double distance = 0.0;
|
||||
int bpos = 0;
|
||||
|
||||
for (int i = 0; i < a->nnz; i++)
|
||||
{
|
||||
int ai = a->indices[i];
|
||||
|
||||
for (int j = bpos; j < b->nnz; j++)
|
||||
{
|
||||
int bi = b->indices[j];
|
||||
|
||||
/* Only update when the same index */
|
||||
if (ai == bi)
|
||||
distance += ax[i] * bx[j];
|
||||
|
||||
/* Update start for next iteration */
|
||||
if (ai >= bi)
|
||||
bpos = j + 1;
|
||||
|
||||
/* Found or passed it */
|
||||
if (bi >= ai)
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
return distance;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the inner product of two sparse vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_inner_product);
|
||||
Datum
|
||||
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));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the negative inner product of two sparse vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_negative_inner_product);
|
||||
Datum
|
||||
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));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the cosine distance between two sparse vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_cosine_distance);
|
||||
Datum
|
||||
sparsevec_cosine_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
|
||||
SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
|
||||
float *ax = SPARSEVEC_VALUES(a);
|
||||
float *bx = SPARSEVEC_VALUES(b);
|
||||
float norma = 0.0;
|
||||
float normb = 0.0;
|
||||
double similarity;
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
similarity = inner_product_internal(a, b);
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < a->nnz; i++)
|
||||
norma += ax[i] * ax[i];
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < b->nnz; i++)
|
||||
normb += bx[i] * bx[i];
|
||||
|
||||
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
|
||||
similarity /= sqrt((double) norma * (double) normb);
|
||||
|
||||
#ifdef _MSC_VER
|
||||
/* /fp:fast may not propagate NaN */
|
||||
if (isnan(similarity))
|
||||
PG_RETURN_FLOAT8(NAN);
|
||||
#endif
|
||||
|
||||
/* Keep in range */
|
||||
if (similarity > 1)
|
||||
similarity = 1.0;
|
||||
else if (similarity < -1)
|
||||
similarity = -1.0;
|
||||
|
||||
PG_RETURN_FLOAT8(1.0 - similarity);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 norm of a sparse vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_norm);
|
||||
Datum
|
||||
sparsevec_norm(PG_FUNCTION_ARGS)
|
||||
{
|
||||
SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
|
||||
float *ax = SPARSEVEC_VALUES(a);
|
||||
double norm = 0.0;
|
||||
|
||||
/* Auto-vectorized */
|
||||
for (int i = 0; i < a->nnz; i++)
|
||||
norm += (double) ax[i] * (double) ax[i];
|
||||
|
||||
PG_RETURN_FLOAT8(sqrt(norm));
|
||||
}
|
||||
@@ -1,25 +0,0 @@
|
||||
#ifndef SPARSEVEC_H
|
||||
#define SPARSEVEC_H
|
||||
|
||||
#define SPARSEVEC_MAX_DIM 100000
|
||||
#define SPARSEVEC_MAX_NNZ 16000
|
||||
|
||||
/* Ensure values are aligned */
|
||||
#define SPARSEVEC_SIZE(_nnz) (offsetof(SparseVector, indices) + MAXALIGN((_nnz) * sizeof(int32)) + (_nnz * sizeof(float)))
|
||||
#define SPARSEVEC_VALUES(x) ((float *) (((char *) (x)) + offsetof(SparseVector, indices) + MAXALIGN((x)->nnz * sizeof(int32))))
|
||||
#define DatumGetSparseVector(x) ((SparseVector *) PG_DETOAST_DATUM(x))
|
||||
#define PG_GETARG_SPARSEVEC_P(x) DatumGetSparseVector(PG_GETARG_DATUM(x))
|
||||
#define PG_RETURN_SPARSEVEC_P(x) PG_RETURN_POINTER(x)
|
||||
|
||||
typedef struct SparseVector
|
||||
{
|
||||
int32 vl_len_; /* varlena header (do not touch directly!) */
|
||||
int32 dim; /* number of dimensions */
|
||||
int32 nnz;
|
||||
int32 unused;
|
||||
int32 indices[FLEXIBLE_ARRAY_MEMBER];
|
||||
} SparseVector;
|
||||
|
||||
SparseVector *InitSparseVector(int dim, int nnz);
|
||||
|
||||
#endif
|
||||
87
src/vector.c
87
src/vector.c
@@ -6,13 +6,11 @@
|
||||
#include "catalog/pg_type.h"
|
||||
#include "common/shortest_dec.h"
|
||||
#include "fmgr.h"
|
||||
#include "halfvec.h"
|
||||
#include "hnsw.h"
|
||||
#include "ivfflat.h"
|
||||
#include "lib/stringinfo.h"
|
||||
#include "libpq/pqformat.h"
|
||||
#include "port.h" /* for strtof() */
|
||||
#include "sparsevec.h"
|
||||
#include "utils/array.h"
|
||||
#include "utils/builtins.h"
|
||||
#include "utils/float.h"
|
||||
@@ -200,8 +198,6 @@ vector_in(PG_FUNCTION_ARGS)
|
||||
|
||||
while (pt != NULL && *stringEnd != ']')
|
||||
{
|
||||
float val;
|
||||
|
||||
if (dim == VECTOR_MAX_DIM)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
@@ -217,23 +213,15 @@ vector_in(PG_FUNCTION_ARGS)
|
||||
errmsg("invalid input syntax for type vector: \"%s\"", lit)));
|
||||
|
||||
/* Use strtof like float4in to avoid a double-rounding problem */
|
||||
errno = 0;
|
||||
val = strtof(pt, &stringEnd);
|
||||
x[dim] = strtof(pt, &stringEnd);
|
||||
CheckElement(x[dim]);
|
||||
dim++;
|
||||
|
||||
if (stringEnd == pt)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
|
||||
errmsg("invalid input syntax for type vector: \"%s\"", lit)));
|
||||
|
||||
/* Check for range error like float4in */
|
||||
if (errno == ERANGE && isinf(val))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
|
||||
errmsg("\"%s\" is out of range for type vector", pt)));
|
||||
|
||||
CheckElement(val);
|
||||
x[dim++] = val;
|
||||
|
||||
while (vector_isspace(*stringEnd))
|
||||
stringEnd++;
|
||||
|
||||
@@ -543,28 +531,6 @@ vector_to_float4(PG_FUNCTION_ARGS)
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert half vector to vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_to_vector);
|
||||
Datum
|
||||
halfvec_to_vector(PG_FUNCTION_ARGS)
|
||||
{
|
||||
HalfVector *vec = PG_GETARG_HALFVEC_P(0);
|
||||
int32 typmod = PG_GETARG_INT32(1);
|
||||
Vector *result;
|
||||
|
||||
CheckDim(vec->dim);
|
||||
CheckExpectedDim(typmod, vec->dim);
|
||||
|
||||
result = InitVector(vec->dim);
|
||||
|
||||
for (int i = 0; i < vec->dim; i++)
|
||||
result->x[i] = HalfToFloat4(vec->x[i]);
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the L2 distance between vectors
|
||||
*/
|
||||
@@ -943,6 +909,30 @@ subvector(PG_FUNCTION_ARGS)
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Concatenate vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_concat);
|
||||
Datum
|
||||
vector_concat(PG_FUNCTION_ARGS)
|
||||
{
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
Vector *result;
|
||||
int dim = a->dim + b->dim;
|
||||
|
||||
CheckDim(dim);
|
||||
result = InitVector(dim);
|
||||
|
||||
for (int i = 0; i < a->dim; i++)
|
||||
result->x[i] = a->x[i];
|
||||
|
||||
for (int i = 0; i < b->dim; i++)
|
||||
result->x[i + a->dim] = b->x[i];
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Internal helper to compare vectors
|
||||
*/
|
||||
@@ -1225,26 +1215,3 @@ vector_avg(PG_FUNCTION_ARGS)
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert sparse vector to dense vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_to_vector);
|
||||
Datum
|
||||
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;
|
||||
float *values = SPARSEVEC_VALUES(svec);
|
||||
|
||||
CheckDim(dim);
|
||||
CheckExpectedDim(typmod, dim);
|
||||
|
||||
result = InitVector(dim);
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
result->x[svec->indices[i] - 1] = values[i];
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
@@ -1,104 +1,64 @@
|
||||
SELECT hamming_distance('111', '111');
|
||||
SELECT hamming_distance(B'111', B'111');
|
||||
hamming_distance
|
||||
------------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('111', '110');
|
||||
SELECT hamming_distance(B'111', B'110');
|
||||
hamming_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('111', '100');
|
||||
SELECT hamming_distance(B'111', B'100');
|
||||
hamming_distance
|
||||
------------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('111', '000');
|
||||
SELECT hamming_distance(B'111', B'000');
|
||||
hamming_distance
|
||||
------------------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('10101010101010101010', '01010101010101010101');
|
||||
hamming_distance
|
||||
------------------
|
||||
20
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('', '');
|
||||
hamming_distance
|
||||
------------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('111', '00');
|
||||
SELECT hamming_distance(B'111', B'00');
|
||||
ERROR: different bit lengths 3 and 2
|
||||
SELECT hamming_distance('111', '000'::varbit(4));
|
||||
hamming_distance
|
||||
------------------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance('111', '0000'::varbit(4));
|
||||
ERROR: different bit lengths 3 and 4
|
||||
SELECT jaccard_distance('1111', '1111');
|
||||
SELECT jaccard_distance(B'1111', B'1111');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '1110');
|
||||
SELECT jaccard_distance(B'1111', B'1110');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.25
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '1100');
|
||||
SELECT jaccard_distance(B'1111', B'1100');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.5
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '1000');
|
||||
SELECT jaccard_distance(B'1111', B'1000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.75
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '0000');
|
||||
SELECT jaccard_distance(B'1111', B'0000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1100', '1000');
|
||||
SELECT jaccard_distance(B'1100', B'1000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.5
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
|
||||
jaccard_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('', '');
|
||||
jaccard_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '000');
|
||||
SELECT jaccard_distance(B'1111', B'000');
|
||||
ERROR: different bit lengths 4 and 3
|
||||
SELECT jaccard_distance('1111', '0000'::varbit(5));
|
||||
jaccard_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance('1111', '00000'::varbit(5));
|
||||
ERROR: different bit lengths 4 and 5
|
||||
|
||||
@@ -46,30 +46,6 @@ SELECT '[1,2,3]'::vector::real[];
|
||||
{1,2,3}
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector::halfvec;
|
||||
halfvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::halfvec::vector;
|
||||
vector
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector::halfvec(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[1,2,3]'::halfvec::vector(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[65520]'::vector::halfvec;
|
||||
ERROR: infinite value not allowed in halfvec
|
||||
SELECT '[1e-8]'::vector::halfvec;
|
||||
halfvec
|
||||
---------
|
||||
[0]
|
||||
(1 row)
|
||||
|
||||
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
|
||||
ERROR: vector cannot have more than 16000 dimensions
|
||||
SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
|
||||
|
||||
@@ -1,15 +1,15 @@
|
||||
CREATE TABLE t (val vector(3), val2 halfvec(3));
|
||||
INSERT INTO t (val, val2) VALUES ('[0,0,0]', '[0,0,0]'), ('[1,2,3]', '[1,2,3]'), ('[1,1,1]', '[1,1,1]'), (NULL, NULL);
|
||||
CREATE TABLE t2 (val vector(3), val2 halfvec(3));
|
||||
CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE TABLE t2 (val vector(3));
|
||||
\copy t TO 'results/data.bin' WITH (FORMAT binary)
|
||||
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
|
||||
SELECT * FROM t2 ORDER BY val;
|
||||
val | val2
|
||||
---------+---------
|
||||
[0,0,0] | [0,0,0]
|
||||
[1,1,1] | [1,1,1]
|
||||
[1,2,3] | [1,2,3]
|
||||
|
|
||||
val
|
||||
---------
|
||||
[0,0,0]
|
||||
[1,1,1]
|
||||
[1,2,3]
|
||||
|
||||
(4 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
@@ -1,146 +0,0 @@
|
||||
SELECT l2_distance('[0,0]'::halfvec, '[3,4]');
|
||||
l2_distance
|
||||
-------------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[0,0]'::halfvec, '[0,1]');
|
||||
l2_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[1,2]'::halfvec, '[3]');
|
||||
ERROR: different halfvec dimensions 2 and 1
|
||||
SELECT '[0,0]'::halfvec <-> '[3,4]';
|
||||
?column?
|
||||
----------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::halfvec, '[3,4]');
|
||||
inner_product
|
||||
---------------
|
||||
11
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::halfvec, '[3]');
|
||||
ERROR: different halfvec dimensions 2 and 1
|
||||
SELECT inner_product('[65504]'::halfvec, '[65504]');
|
||||
inner_product
|
||||
---------------
|
||||
4290774016
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2]'::halfvec <#> '[3,4]';
|
||||
?column?
|
||||
----------
|
||||
-11
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[2,4]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[0,0]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[1,1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,0]'::halfvec, '[0,2]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[-1,-1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[3]');
|
||||
ERROR: different halfvec dimensions 2 and 1
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[1.1,1.1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[-1.1,-1.1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2]'::halfvec <=> '[2,4]';
|
||||
?column?
|
||||
----------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::halfvec, '[3,4]');
|
||||
l1_distance
|
||||
-------------
|
||||
7
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::halfvec, '[0,1]');
|
||||
l1_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[1,2]'::halfvec, '[3]');
|
||||
ERROR: different halfvec dimensions 2 and 1
|
||||
SELECT quantize_binary('[1,0,-1]'::halfvec);
|
||||
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]'::halfvec);
|
||||
quantize_binary
|
||||
-----------------
|
||||
01001110101
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2);
|
||||
subvector
|
||||
-----------
|
||||
[3,4]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 9);
|
||||
subvector
|
||||
-----------
|
||||
[3,4,5]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 0);
|
||||
ERROR: halfvec must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, -1);
|
||||
ERROR: halfvec must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 2);
|
||||
ERROR: halfvec must have at least 1 dimension
|
||||
@@ -1,151 +0,0 @@
|
||||
SELECT '[1,2,3]'::halfvec;
|
||||
halfvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[-1,-2,-3]'::halfvec;
|
||||
halfvec
|
||||
------------
|
||||
[-1,-2,-3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1.,2.,3.]'::halfvec;
|
||||
halfvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT ' [ 1, 2 , 3 ] '::halfvec;
|
||||
halfvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1.23456]'::halfvec;
|
||||
halfvec
|
||||
------------
|
||||
[1.234375]
|
||||
(1 row)
|
||||
|
||||
SELECT '[hello,1]'::halfvec;
|
||||
ERROR: invalid input syntax for type halfvec: "[hello,1]"
|
||||
LINE 1: SELECT '[hello,1]'::halfvec;
|
||||
^
|
||||
SELECT '[NaN,1]'::halfvec;
|
||||
ERROR: NaN not allowed in halfvec
|
||||
LINE 1: SELECT '[NaN,1]'::halfvec;
|
||||
^
|
||||
SELECT '[Infinity,1]'::halfvec;
|
||||
ERROR: infinite value not allowed in halfvec
|
||||
LINE 1: SELECT '[Infinity,1]'::halfvec;
|
||||
^
|
||||
SELECT '[-Infinity,1]'::halfvec;
|
||||
ERROR: infinite value not allowed in halfvec
|
||||
LINE 1: SELECT '[-Infinity,1]'::halfvec;
|
||||
^
|
||||
SELECT '[65519,-65519]'::halfvec;
|
||||
halfvec
|
||||
----------------
|
||||
[65504,-65504]
|
||||
(1 row)
|
||||
|
||||
SELECT '[65520,-65520]'::halfvec;
|
||||
ERROR: "65520" is out of range for type halfvec
|
||||
LINE 1: SELECT '[65520,-65520]'::halfvec;
|
||||
^
|
||||
SELECT '[1e-8,-1e-8]'::halfvec;
|
||||
ERROR: "1e-8" is out of range for type halfvec
|
||||
LINE 1: SELECT '[1e-8,-1e-8]'::halfvec;
|
||||
^
|
||||
SELECT '[4e38,1]'::halfvec;
|
||||
ERROR: "4e38" is out of range for type halfvec
|
||||
LINE 1: SELECT '[4e38,1]'::halfvec;
|
||||
^
|
||||
SELECT '[1e-46,1]'::halfvec;
|
||||
ERROR: "1e-46" is out of range for type halfvec
|
||||
LINE 1: SELECT '[1e-46,1]'::halfvec;
|
||||
^
|
||||
SELECT '[1,2,3'::halfvec;
|
||||
ERROR: malformed halfvec literal: "[1,2,3"
|
||||
LINE 1: SELECT '[1,2,3'::halfvec;
|
||||
^
|
||||
DETAIL: Unexpected end of input.
|
||||
SELECT '[1,2,3]9'::halfvec;
|
||||
ERROR: malformed halfvec literal: "[1,2,3]9"
|
||||
LINE 1: SELECT '[1,2,3]9'::halfvec;
|
||||
^
|
||||
DETAIL: Junk after closing right brace.
|
||||
SELECT '1,2,3'::halfvec;
|
||||
ERROR: malformed halfvec literal: "1,2,3"
|
||||
LINE 1: SELECT '1,2,3'::halfvec;
|
||||
^
|
||||
DETAIL: Vector contents must start with "[".
|
||||
SELECT ''::halfvec;
|
||||
ERROR: malformed halfvec literal: ""
|
||||
LINE 1: SELECT ''::halfvec;
|
||||
^
|
||||
DETAIL: Vector contents must start with "[".
|
||||
SELECT '['::halfvec;
|
||||
ERROR: malformed halfvec literal: "["
|
||||
LINE 1: SELECT '['::halfvec;
|
||||
^
|
||||
DETAIL: Unexpected end of input.
|
||||
SELECT '[,'::halfvec;
|
||||
ERROR: malformed halfvec literal: "[,"
|
||||
LINE 1: SELECT '[,'::halfvec;
|
||||
^
|
||||
DETAIL: Unexpected end of input.
|
||||
SELECT '[]'::halfvec;
|
||||
ERROR: halfvec must have at least 1 dimension
|
||||
LINE 1: SELECT '[]'::halfvec;
|
||||
^
|
||||
SELECT '[1,]'::halfvec;
|
||||
ERROR: invalid input syntax for type halfvec: "[1,]"
|
||||
LINE 1: SELECT '[1,]'::halfvec;
|
||||
^
|
||||
SELECT '[1a]'::halfvec;
|
||||
ERROR: invalid input syntax for type halfvec: "[1a]"
|
||||
LINE 1: SELECT '[1a]'::halfvec;
|
||||
^
|
||||
SELECT '[1,,3]'::halfvec;
|
||||
ERROR: malformed halfvec literal: "[1,,3]"
|
||||
LINE 1: SELECT '[1,,3]'::halfvec;
|
||||
^
|
||||
SELECT '[1, ,3]'::halfvec;
|
||||
ERROR: invalid input syntax for type halfvec: "[1, ,3]"
|
||||
LINE 1: SELECT '[1, ,3]'::halfvec;
|
||||
^
|
||||
SELECT '[1,2,3]'::halfvec(3);
|
||||
halfvec
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::halfvec(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[1,2,3]'::halfvec(3, 2);
|
||||
ERROR: invalid type modifier
|
||||
LINE 1: SELECT '[1,2,3]'::halfvec(3, 2);
|
||||
^
|
||||
SELECT '[1,2,3]'::halfvec('a');
|
||||
ERROR: invalid input syntax for type integer: "a"
|
||||
LINE 1: SELECT '[1,2,3]'::halfvec('a');
|
||||
^
|
||||
SELECT '[1,2,3]'::halfvec(0);
|
||||
ERROR: dimensions for type halfvec must be at least 1
|
||||
LINE 1: SELECT '[1,2,3]'::halfvec(0);
|
||||
^
|
||||
SELECT '[1,2,3]'::halfvec(16001);
|
||||
ERROR: dimensions for type halfvec cannot exceed 16000
|
||||
LINE 1: SELECT '[1,2,3]'::halfvec(16001);
|
||||
^
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::halfvec[]);
|
||||
unnest
|
||||
---------
|
||||
[1,2,3]
|
||||
[4,5,6]
|
||||
(2 rows)
|
||||
|
||||
SELECT '{"[1,2,3]"}'::halfvec(2)[];
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
@@ -1,26 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_cosine_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,1,1]
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
(3 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,21 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_ip_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,4]
|
||||
[1,2,3]
|
||||
[1,1,1]
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::halfvec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,33 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_l2_ops);
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
---------
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
[1,1,1]
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::halfvec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM t;
|
||||
count
|
||||
-------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -19,9 +19,3 @@ SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
-- TODO move
|
||||
CREATE TABLE t (val varbit(3));
|
||||
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
|
||||
ERROR: type not supported for hnsw index
|
||||
CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
|
||||
DROP TABLE t;
|
||||
@@ -1,26 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
SELECT * FROM t ORDER BY val <=> '{1:3,2:3,3:3}/3';
|
||||
val
|
||||
-----------------
|
||||
{1:1,2:1,3:1}/3
|
||||
{1:1,2:2,3:3}/3
|
||||
{1:1,2:2,3:4}/3
|
||||
(3 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
|
||||
count
|
||||
-------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,21 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
|
||||
val
|
||||
-----------------
|
||||
{1:1,2:2,3:4}/3
|
||||
{1:1,2:2,3:3}/3
|
||||
{1:1,2:1,3:1}/3
|
||||
{}/3
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,43 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/3';
|
||||
val
|
||||
-----------------
|
||||
{1:1,2:2,3:3}/3
|
||||
{1:1,2:2,3:4}/3
|
||||
{1:1,2:1,3:1}/3
|
||||
{}/3
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM t;
|
||||
count
|
||||
-------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/3';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
-- TODO move
|
||||
CREATE TABLE t (val sparsevec(1001));
|
||||
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
|
||||
TRUNCATE t;
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
|
||||
ERROR: sparsevec cannot have more than 1000 non-zero elements for hnsw index
|
||||
DROP TABLE t;
|
||||
@@ -1,62 +0,0 @@
|
||||
SELECT l2_distance('{}/2'::sparsevec, '{1:3,2:4}/2');
|
||||
l2_distance
|
||||
-------------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('{}/2'::sparsevec, '{2:1}/2');
|
||||
l2_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT '{}/2'::sparsevec <-> '{1:3,2:4}/2';
|
||||
?column?
|
||||
----------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
|
||||
inner_product
|
||||
---------------
|
||||
10
|
||||
(1 row)
|
||||
|
||||
SELECT sparsevec_negative_inner_product('{1:1,2:2}/2', '{1:2,2:4}/2');
|
||||
sparsevec_negative_inner_product
|
||||
----------------------------------
|
||||
-10
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{}/2');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:-1,2:-1}/2');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{1:2}/2'::sparsevec, '{2:2}/2');
|
||||
cosine_distance
|
||||
-----------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{1:2}/2'::sparsevec, '{1:1}/3');
|
||||
ERROR: different sparsevec dimensions 2 and 3
|
||||
@@ -1,62 +0,0 @@
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec;
|
||||
sparsevec
|
||||
-----------------
|
||||
{1:1.5,3:3.5}/5
|
||||
(1 row)
|
||||
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector;
|
||||
vector
|
||||
-----------------
|
||||
[1.5,0,3.5,0,0]
|
||||
(1 row)
|
||||
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector(5);
|
||||
vector
|
||||
-----------------
|
||||
[1.5,0,3.5,0,0]
|
||||
(1 row)
|
||||
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector(4);
|
||||
ERROR: expected 4 dimensions, not 5
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
sparsevec
|
||||
-----------------
|
||||
{2:1.5,4:3.5}/5
|
||||
(1 row)
|
||||
|
||||
SELECT '{1:0,2:1,3:0}/3'::sparsevec;
|
||||
sparsevec
|
||||
-----------
|
||||
{2:1}/3
|
||||
(1 row)
|
||||
|
||||
SELECT '{2:1,1:1}/2'::sparsevec;
|
||||
ERROR: indexes must be in ascending order
|
||||
LINE 1: SELECT '{2:1,1:1}/2'::sparsevec;
|
||||
^
|
||||
SELECT '{}/5'::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 '{0:1}/1'::sparsevec;
|
||||
ERROR: index "0" is out of range for type sparsevec
|
||||
LINE 1: SELECT '{0:1}/1'::sparsevec;
|
||||
^
|
||||
SELECT '{2:1}/1'::sparsevec;
|
||||
ERROR: index must be less than or equal to dimensions
|
||||
LINE 1: SELECT '{2:1}/1'::sparsevec;
|
||||
^
|
||||
SELECT '{}/1'::sparsevec(2);
|
||||
ERROR: expected 2 dimensions, not 1
|
||||
@@ -24,6 +24,14 @@ SELECT '[1e37]'::vector * '[1e37]';
|
||||
ERROR: value out of range: overflow
|
||||
SELECT '[1e-37]'::vector * '[1e-37]';
|
||||
ERROR: value out of range: underflow
|
||||
SELECT '[1,2,3]'::vector || '[4,5]'::vector;
|
||||
?column?
|
||||
-------------
|
||||
[1,2,3,4,5]
|
||||
(1 row)
|
||||
|
||||
SELECT array_fill(0, ARRAY[16000])::vector || '[1]'::vector;
|
||||
ERROR: vector cannot have more than 16000 dimensions
|
||||
SELECT '[1,2,3]'::vector < '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
@@ -272,47 +280,47 @@ SELECT l1_distance('[3e38]'::vector, '[-3e38]');
|
||||
Infinity
|
||||
(1 row)
|
||||
|
||||
SELECT quantize_binary('[1,0,-1]'::vector);
|
||||
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]'::vector);
|
||||
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 subvector('[1,2,3,4,5]'::vector, 1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]', 1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, 2);
|
||||
subvector
|
||||
-----------
|
||||
[3,4]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]', -1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 9);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, 9);
|
||||
subvector
|
||||
-----------
|
||||
[3,4,5]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 1, 0);
|
||||
SELECT subvector('[1,2,3,4,5]', 1, 0);
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, -1);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, -1);
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -1, 2);
|
||||
SELECT subvector('[1,2,3,4,5]', -1, 2);
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
|
||||
avg
|
||||
|
||||
@@ -63,11 +63,11 @@ SELECT '[1.5e-38,-1.5e-38]'::vector;
|
||||
(1 row)
|
||||
|
||||
SELECT '[4e38,1]'::vector;
|
||||
ERROR: "4e38" is out of range for type vector
|
||||
ERROR: infinite value not allowed in vector
|
||||
LINE 1: SELECT '[4e38,1]'::vector;
|
||||
^
|
||||
SELECT '[-4e38,1]'::vector;
|
||||
ERROR: "-4e38" is out of range for type vector
|
||||
ERROR: infinite value not allowed in vector
|
||||
LINE 1: SELECT '[-4e38,1]'::vector;
|
||||
^
|
||||
SELECT '[1e-46,1]'::vector;
|
||||
|
||||
@@ -1,21 +1,13 @@
|
||||
SELECT hamming_distance('111', '111');
|
||||
SELECT hamming_distance('111', '110');
|
||||
SELECT hamming_distance('111', '100');
|
||||
SELECT hamming_distance('111', '000');
|
||||
SELECT hamming_distance('10101010101010101010', '01010101010101010101');
|
||||
SELECT hamming_distance('', '');
|
||||
SELECT hamming_distance('111', '00');
|
||||
SELECT hamming_distance('111', '000'::varbit(4));
|
||||
SELECT hamming_distance('111', '0000'::varbit(4));
|
||||
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('1111', '1111');
|
||||
SELECT jaccard_distance('1111', '1110');
|
||||
SELECT jaccard_distance('1111', '1100');
|
||||
SELECT jaccard_distance('1111', '1000');
|
||||
SELECT jaccard_distance('1111', '0000');
|
||||
SELECT jaccard_distance('1100', '1000');
|
||||
SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
|
||||
SELECT jaccard_distance('', '');
|
||||
SELECT jaccard_distance('1111', '000');
|
||||
SELECT jaccard_distance('1111', '0000'::varbit(5));
|
||||
SELECT jaccard_distance('1111', '00000'::varbit(5));
|
||||
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');
|
||||
|
||||
@@ -10,12 +10,6 @@ SELECT '{-Infinity}'::real[]::vector;
|
||||
SELECT '{}'::real[]::vector;
|
||||
SELECT '{{1}}'::real[]::vector;
|
||||
SELECT '[1,2,3]'::vector::real[];
|
||||
SELECT '[1,2,3]'::vector::halfvec;
|
||||
SELECT '[1,2,3]'::halfvec::vector;
|
||||
SELECT '[1,2,3]'::vector::halfvec(2);
|
||||
SELECT '[1,2,3]'::halfvec::vector(2);
|
||||
SELECT '[65520]'::vector::halfvec;
|
||||
SELECT '[1e-8]'::vector::halfvec;
|
||||
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
|
||||
SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
|
||||
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
CREATE TABLE t (val vector(3), val2 halfvec(3));
|
||||
INSERT INTO t (val, val2) VALUES ('[0,0,0]', '[0,0,0]'), ('[1,2,3]', '[1,2,3]'), ('[1,1,1]', '[1,1,1]'), (NULL, NULL);
|
||||
CREATE TABLE t (val vector(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
|
||||
CREATE TABLE t2 (val vector(3), val2 halfvec(3));
|
||||
CREATE TABLE t2 (val vector(3));
|
||||
|
||||
\copy t TO 'results/data.bin' WITH (FORMAT binary)
|
||||
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
|
||||
|
||||
@@ -1,34 +0,0 @@
|
||||
SELECT l2_distance('[0,0]'::halfvec, '[3,4]');
|
||||
SELECT l2_distance('[0,0]'::halfvec, '[0,1]');
|
||||
SELECT l2_distance('[1,2]'::halfvec, '[3]');
|
||||
SELECT '[0,0]'::halfvec <-> '[3,4]';
|
||||
|
||||
SELECT inner_product('[1,2]'::halfvec, '[3,4]');
|
||||
SELECT inner_product('[1,2]'::halfvec, '[3]');
|
||||
SELECT inner_product('[65504]'::halfvec, '[65504]');
|
||||
SELECT '[1,2]'::halfvec <#> '[3,4]';
|
||||
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[2,4]');
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[0,0]');
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[1,1]');
|
||||
SELECT cosine_distance('[1,0]'::halfvec, '[0,2]');
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[-1,-1]');
|
||||
SELECT cosine_distance('[1,2]'::halfvec, '[3]');
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[1.1,1.1]');
|
||||
SELECT cosine_distance('[1,1]'::halfvec, '[-1.1,-1.1]');
|
||||
SELECT '[1,2]'::halfvec <=> '[2,4]';
|
||||
|
||||
SELECT l1_distance('[0,0]'::halfvec, '[3,4]');
|
||||
SELECT l1_distance('[0,0]'::halfvec, '[0,1]');
|
||||
SELECT l1_distance('[1,2]'::halfvec, '[3]');
|
||||
|
||||
SELECT quantize_binary('[1,0,-1]'::halfvec);
|
||||
SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::halfvec);
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 9);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 0);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, -1);
|
||||
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 2);
|
||||
@@ -1,35 +0,0 @@
|
||||
SELECT '[1,2,3]'::halfvec;
|
||||
SELECT '[-1,-2,-3]'::halfvec;
|
||||
SELECT '[1.,2.,3.]'::halfvec;
|
||||
SELECT ' [ 1, 2 , 3 ] '::halfvec;
|
||||
SELECT '[1.23456]'::halfvec;
|
||||
SELECT '[hello,1]'::halfvec;
|
||||
SELECT '[NaN,1]'::halfvec;
|
||||
SELECT '[Infinity,1]'::halfvec;
|
||||
SELECT '[-Infinity,1]'::halfvec;
|
||||
SELECT '[65519,-65519]'::halfvec;
|
||||
SELECT '[65520,-65520]'::halfvec;
|
||||
SELECT '[1e-8,-1e-8]'::halfvec;
|
||||
SELECT '[4e38,1]'::halfvec;
|
||||
SELECT '[1e-46,1]'::halfvec;
|
||||
SELECT '[1,2,3'::halfvec;
|
||||
SELECT '[1,2,3]9'::halfvec;
|
||||
SELECT '1,2,3'::halfvec;
|
||||
SELECT ''::halfvec;
|
||||
SELECT '['::halfvec;
|
||||
SELECT '[,'::halfvec;
|
||||
SELECT '[]'::halfvec;
|
||||
SELECT '[1,]'::halfvec;
|
||||
SELECT '[1a]'::halfvec;
|
||||
SELECT '[1,,3]'::halfvec;
|
||||
SELECT '[1, ,3]'::halfvec;
|
||||
|
||||
SELECT '[1,2,3]'::halfvec(3);
|
||||
SELECT '[1,2,3]'::halfvec(2);
|
||||
SELECT '[1,2,3]'::halfvec(3, 2);
|
||||
SELECT '[1,2,3]'::halfvec('a');
|
||||
SELECT '[1,2,3]'::halfvec(0);
|
||||
SELECT '[1,2,3]'::halfvec(16001);
|
||||
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::halfvec[]);
|
||||
SELECT '{"[1,2,3]"}'::halfvec(2)[];
|
||||
@@ -1,13 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_cosine_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,12 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_ip_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::halfvec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,16 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val halfvec(3));
|
||||
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val halfvec_l2_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('[1,2,4]');
|
||||
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::halfvec)) t2;
|
||||
SELECT COUNT(*) FROM t;
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -10,9 +10,3 @@ SELECT * FROM t ORDER BY val <~> B'111';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- TODO move
|
||||
CREATE TABLE t (val varbit(3));
|
||||
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
|
||||
CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
|
||||
DROP TABLE t;
|
||||
@@ -1,13 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
|
||||
SELECT * FROM t ORDER BY val <=> '{1:3,2:3,3:3}/3';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,12 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
|
||||
SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,25 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec(3));
|
||||
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
|
||||
|
||||
SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/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 <-> '{1:3,2:3,3:3}/3';
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- TODO move
|
||||
CREATE TABLE t (val sparsevec(1001));
|
||||
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
TRUNCATE t;
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
|
||||
DROP TABLE t;
|
||||
@@ -1,13 +0,0 @@
|
||||
SELECT l2_distance('{}/2'::sparsevec, '{1:3,2:4}/2');
|
||||
SELECT l2_distance('{}/2'::sparsevec, '{2:1}/2');
|
||||
SELECT '{}/2'::sparsevec <-> '{1:3,2:4}/2';
|
||||
|
||||
SELECT inner_product('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
|
||||
SELECT sparsevec_negative_inner_product('{1:1,2:2}/2', '{1:2,2:4}/2');
|
||||
|
||||
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
|
||||
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{}/2');
|
||||
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:-1,2:-1}/2');
|
||||
SELECT cosine_distance('{1:2}/2'::sparsevec, '{2:2}/2');
|
||||
SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
|
||||
SELECT cosine_distance('{1:2}/2'::sparsevec, '{1:1}/3');
|
||||
@@ -1,19 +0,0 @@
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec;
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector;
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector(5);
|
||||
SELECT '{1:1.5,3:3.5}/5'::sparsevec::vector(4);
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
|
||||
SELECT '{1:0,2:1,3:0}/3'::sparsevec;
|
||||
|
||||
SELECT '{2:1,1:1}/2'::sparsevec;
|
||||
|
||||
SELECT '{}/5'::sparsevec;
|
||||
SELECT '{}/-1'::sparsevec;
|
||||
SELECT '{}/100001'::sparsevec;
|
||||
SELECT '{}/16001'::sparsevec::vector;
|
||||
|
||||
SELECT '{0:1}/1'::sparsevec;
|
||||
SELECT '{2:1}/1'::sparsevec;
|
||||
|
||||
SELECT '{}/1'::sparsevec(2);
|
||||
@@ -6,6 +6,9 @@ SELECT '[1,2,3]'::vector * '[4,5,6]';
|
||||
SELECT '[1e37]'::vector * '[1e37]';
|
||||
SELECT '[1e-37]'::vector * '[1e-37]';
|
||||
|
||||
SELECT '[1,2,3]'::vector || '[4,5]'::vector;
|
||||
SELECT array_fill(0, ARRAY[16000])::vector || '[1]'::vector;
|
||||
|
||||
SELECT '[1,2,3]'::vector < '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector < '[1,2]';
|
||||
SELECT '[1,2,3]'::vector <= '[1,2,3]';
|
||||
@@ -58,16 +61,16 @@ 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]'::vector);
|
||||
SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::vector);
|
||||
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 subvector('[1,2,3,4,5]'::vector, 1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 9);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 1, 0);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, -1);
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -1, 2);
|
||||
SELECT subvector('[1,2,3,4,5]', 1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, 2);
|
||||
SELECT subvector('[1,2,3,4,5]', -1, 3);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, 9);
|
||||
SELECT subvector('[1,2,3,4,5]', 1, 0);
|
||||
SELECT subvector('[1,2,3,4,5]', 3, -1);
|
||||
SELECT subvector('[1,2,3,4,5]', -1, 2);
|
||||
|
||||
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;
|
||||
|
||||
@@ -86,7 +86,7 @@ foreach (@queries)
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
test_recall(0.18, $limit, "before vacuum");
|
||||
test_recall(0.19, $limit, "before vacuum");
|
||||
test_recall(0.95, 100, "before vacuum");
|
||||
|
||||
# TODO Test concurrent inserts with vacuum
|
||||
|
||||
@@ -99,7 +99,7 @@ for my $i (0 .. $#operators)
|
||||
));
|
||||
|
||||
# Test approximate results
|
||||
my $min = $operator eq "<\%>" ? 0.95 : 0.98;
|
||||
my $min = $operator eq "<\%>" ? 0.96 : 0.98;
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
@@ -1,132 +0,0 @@
|
||||
use strict;
|
||||
use warnings;
|
||||
use PostgresNode;
|
||||
use TestLib;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
my $dim = 10;
|
||||
my $array_sql = join(",", ('random()') x $dim);
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $operator) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan/);
|
||||
|
||||
for my $i (0 .. $#queries)
|
||||
{
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
|
||||
));
|
||||
my @actual_ids = split("\n", $actual);
|
||||
my %actual_set = map { $_ => 1 } @actual_ids;
|
||||
|
||||
my @expected_ids = split("\n", $expected[$i]);
|
||||
|
||||
foreach (@expected_ids)
|
||||
{
|
||||
if (exists($actual_set{$_}))
|
||||
{
|
||||
$correct++;
|
||||
}
|
||||
$total++;
|
||||
}
|
||||
}
|
||||
|
||||
cmp_ok($correct / $total, ">=", $min, $operator);
|
||||
}
|
||||
|
||||
# Initialize node
|
||||
$node = get_new_node('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v halfvec($dim));");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my @r = ();
|
||||
for (1 .. $dim)
|
||||
{
|
||||
push(@r, rand());
|
||||
}
|
||||
push(@queries, "[" . join(",", @r) . "]");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<->", "<#>", "<=>");
|
||||
my @opclasses = ("halfvec_l2_ops", "halfvec_ip_ops", "halfvec_cosine_ops");
|
||||
|
||||
for my $i (0 .. $#operators)
|
||||
{
|
||||
my $operator = $operators[$i];
|
||||
my $opclass = $opclasses[$i];
|
||||
|
||||
# Get exact results
|
||||
@expected = ();
|
||||
foreach (@queries)
|
||||
{
|
||||
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
# Build index serially
|
||||
$node->safe_psql("postgres", qq(
|
||||
SET max_parallel_maintenance_workers = 0;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
|
||||
# Test approximate results
|
||||
my $min = $operator eq "<#>" ? 0.95 : 0.98;
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel in memory
|
||||
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
SET client_min_messages = DEBUG;
|
||||
SET min_parallel_table_scan_size = 1;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
|
||||
# Test approximate results
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel on disk
|
||||
# Set parallel_workers on table to use workers with low maintenance_work_mem
|
||||
($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
ALTER TABLE tst SET (parallel_workers = 2);
|
||||
SET client_min_messages = DEBUG;
|
||||
SET maintenance_work_mem = '4MB';
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
ALTER TABLE tst RESET (parallel_workers);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
}
|
||||
|
||||
done_testing();
|
||||
@@ -1,128 +0,0 @@
|
||||
use strict;
|
||||
use warnings;
|
||||
use PostgresNode;
|
||||
use TestLib;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $operator) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
|
||||
));
|
||||
like($explain, qr/Index Scan/);
|
||||
|
||||
for my $i (0 .. $#queries)
|
||||
{
|
||||
my $actual = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
|
||||
));
|
||||
my @actual_ids = split("\n", $actual);
|
||||
my %actual_set = map { $_ => 1 } @actual_ids;
|
||||
|
||||
my @expected_ids = split("\n", $expected[$i]);
|
||||
|
||||
foreach (@expected_ids)
|
||||
{
|
||||
if (exists($actual_set{$_}))
|
||||
{
|
||||
$correct++;
|
||||
}
|
||||
$total++;
|
||||
}
|
||||
}
|
||||
|
||||
cmp_ok($correct / $total, ">=", $min, $operator);
|
||||
}
|
||||
|
||||
# Initialize node
|
||||
$node = get_new_node('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v sparsevec(3));");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()]::vector::sparsevec FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my $r1 = rand();
|
||||
my $r2 = rand();
|
||||
my $r3 = rand();
|
||||
push(@queries, "{1:$r1,2:$r2,3:$r3}/3");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<->", "<#>", "<=>");
|
||||
my @opclasses = ("sparsevec_l2_ops", "sparsevec_ip_ops", "sparsevec_cosine_ops");
|
||||
|
||||
for my $i (0 .. $#operators)
|
||||
{
|
||||
my $operator = $operators[$i];
|
||||
my $opclass = $opclasses[$i];
|
||||
|
||||
# Get exact results
|
||||
@expected = ();
|
||||
foreach (@queries)
|
||||
{
|
||||
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
# Build index serially
|
||||
$node->safe_psql("postgres", qq(
|
||||
SET max_parallel_maintenance_workers = 0;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
|
||||
# Test approximate results
|
||||
my $min = $operator eq "<#>" ? 0.80 : 0.99;
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel in memory
|
||||
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
SET client_min_messages = DEBUG;
|
||||
SET min_parallel_table_scan_size = 1;
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
|
||||
# Test approximate results
|
||||
test_recall($min, $operator);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
|
||||
# Build index in parallel on disk
|
||||
# Set parallel_workers on table to use workers with low maintenance_work_mem
|
||||
($ret, $stdout, $stderr) = $node->psql("postgres", qq(
|
||||
ALTER TABLE tst SET (parallel_workers = 2);
|
||||
SET client_min_messages = DEBUG;
|
||||
SET maintenance_work_mem = '4MB';
|
||||
CREATE INDEX idx ON tst USING hnsw (v $opclass);
|
||||
ALTER TABLE tst RESET (parallel_workers);
|
||||
));
|
||||
is($ret, 0, $stderr);
|
||||
like($stderr, qr/using \d+ parallel workers/);
|
||||
like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
}
|
||||
|
||||
done_testing();
|
||||
Reference in New Issue
Block a user