mirror of
https://github.com/pgvector/pgvector.git
synced 2026-07-22 03:57:34 +08:00
Compare commits
1 Commits
sparsevec-
...
hnsw-datum
| Author | SHA1 | Date | |
|---|---|---|---|
|
c8e4bf1c3c |
46
.github/workflows/build.yml
vendored
46
.github/workflows/build.yml
vendored
@@ -28,7 +28,7 @@ jobs:
|
||||
dev-files: true
|
||||
- run: make
|
||||
env:
|
||||
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
|
||||
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
|
||||
- run: |
|
||||
export PG_CONFIG=`which pg_config`
|
||||
sudo --preserve-env=PG_CONFIG make install
|
||||
@@ -40,43 +40,27 @@ jobs:
|
||||
sudo apt-get install libipc-run-perl
|
||||
- run: make prove_installcheck
|
||||
mac:
|
||||
runs-on: ${{ matrix.os }}
|
||||
runs-on: macos-latest
|
||||
if: ${{ !startsWith(github.ref_name, 'windows') }}
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
include:
|
||||
- postgres: 16
|
||||
os: macos-14
|
||||
- postgres: 14
|
||||
os: macos-12
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres@v1
|
||||
with:
|
||||
postgres-version: ${{ matrix.postgres }}
|
||||
postgres-version: 14
|
||||
- run: make
|
||||
env:
|
||||
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
|
||||
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
|
||||
- run: make install
|
||||
- run: make installcheck
|
||||
- if: ${{ failure() }}
|
||||
run: cat regression.diffs
|
||||
# Homebrew Postgres does not enable TAP tests, so need to download
|
||||
- run: |
|
||||
brew install cpanm
|
||||
cpanm --notest IPC::Run
|
||||
wget -q https://github.com/postgres/postgres/archive/refs/tags/$TAG.tar.gz
|
||||
tar xf $TAG.tar.gz
|
||||
mv postgres-$TAG postgres
|
||||
env:
|
||||
TAG: ${{ matrix.postgres == 16 && 'REL_16_2' || 'REL_14_11' }}
|
||||
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
|
||||
env:
|
||||
PERL5LIB: /Users/runner/perl5/lib/perl5
|
||||
- run: make clean && $(brew --prefix llvm@15)/bin/scan-build --status-bugs make
|
||||
env:
|
||||
PG_CFLAGS: -DUSE_ASSERT_CHECKING
|
||||
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_10.tar.gz
|
||||
tar xf REL_14_10.tar.gz
|
||||
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_10/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
|
||||
- run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make PG_CFLAGS="-DUSE_ASSERT_CHECKING"
|
||||
windows:
|
||||
runs-on: windows-latest
|
||||
if: ${{ !startsWith(github.ref_name, 'mac') }}
|
||||
@@ -87,7 +71,6 @@ jobs:
|
||||
postgres-version: 14
|
||||
- run: |
|
||||
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
|
||||
cd %TEMP% && ^
|
||||
nmake /NOLOGO /F Makefile.win && ^
|
||||
nmake /NOLOGO /F Makefile.win install && ^
|
||||
nmake /NOLOGO /F Makefile.win installcheck && ^
|
||||
@@ -114,15 +97,4 @@ jobs:
|
||||
sudo -u postgres make installcheck
|
||||
sudo -u postgres make prove_installcheck
|
||||
env:
|
||||
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
|
||||
valgrind:
|
||||
if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
|
||||
runs-on: ubuntu-latest
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
- uses: ankane/setup-postgres-valgrind@v1
|
||||
with:
|
||||
postgres-version: 16
|
||||
- run: make
|
||||
- run: sudo --preserve-env=PG_CONFIG make install
|
||||
- run: make installcheck
|
||||
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
|
||||
|
||||
19
CHANGELOG.md
19
CHANGELOG.md
@@ -1,23 +1,6 @@
|
||||
## 0.7.0 (unreleased)
|
||||
|
||||
- Added `halfvec` type
|
||||
- Added `sparsevec` type
|
||||
- Added support for `bit` vectors to HNSW
|
||||
- Added `hamming_distance` function
|
||||
- Added `jaccard_distance` function
|
||||
- Added `quantize_binary` function
|
||||
- Added `subvector` function
|
||||
- Updated comparison operators to support vectors with different dimensions
|
||||
|
||||
## 0.6.2 (2024-03-18)
|
||||
|
||||
- Reduced lock contention with parallel HNSW index builds
|
||||
|
||||
## 0.6.1 (2024-03-04)
|
||||
## 0.6.1 (unreleased)
|
||||
|
||||
- Fixed error with `ANALYZE` and vectors with different dimensions
|
||||
- Fixed segmentation fault with `shared_preload_libraries`
|
||||
- Fixed vector subtraction being marked as commutative
|
||||
|
||||
## 0.6.0 (2024-01-29)
|
||||
|
||||
|
||||
2
LICENSE
2
LICENSE
@@ -1,4 +1,4 @@
|
||||
Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
|
||||
Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
|
||||
|
||||
Portions Copyright (c) 1994, The Regents of the University of California
|
||||
|
||||
|
||||
@@ -2,7 +2,7 @@
|
||||
"name": "vector",
|
||||
"abstract": "Open-source vector similarity search for Postgres",
|
||||
"description": "Supports L2 distance, inner product, and cosine distance",
|
||||
"version": "0.6.2",
|
||||
"version": "0.6.0",
|
||||
"maintainer": [
|
||||
"Andrew Kane <andrew@ankane.org>"
|
||||
],
|
||||
@@ -20,7 +20,7 @@
|
||||
"vector": {
|
||||
"file": "sql/vector.sql",
|
||||
"docfile": "README.md",
|
||||
"version": "0.6.2",
|
||||
"version": "0.6.0",
|
||||
"abstract": "Open-source vector similarity search for Postgres"
|
||||
}
|
||||
},
|
||||
|
||||
14
Makefile
14
Makefile
@@ -1,10 +1,10 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.6.2
|
||||
EXTVERSION = 0.6.0
|
||||
|
||||
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/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))
|
||||
@@ -12,7 +12,7 @@ REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
|
||||
|
||||
OPTFLAGS = -march=native
|
||||
|
||||
# Mac ARM doesn't always support -march=native
|
||||
# Mac ARM doesn't support -march=native
|
||||
ifeq ($(shell uname -s), Darwin)
|
||||
ifeq ($(shell uname -p), arm)
|
||||
# no difference with -march=armv8.5-a
|
||||
@@ -71,9 +71,11 @@ PG_MAJOR ?= 16
|
||||
.PHONY: docker
|
||||
|
||||
docker:
|
||||
docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
|
||||
docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
|
||||
docker build --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
|
||||
|
||||
.PHONY: docker-release
|
||||
|
||||
docker-release:
|
||||
docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
|
||||
docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
|
||||
docker buildx build --push --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
|
||||
|
||||
@@ -1,10 +1,10 @@
|
||||
EXTENSION = vector
|
||||
EXTVERSION = 0.6.2
|
||||
EXTVERSION = 0.6.0
|
||||
|
||||
OBJS = src\bitvector.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
|
||||
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
|
||||
OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
|
||||
HEADERS = src\vector.h
|
||||
|
||||
REGRESS = bit_functions btree cast copy halfvec_functions halfvec_input hnsw_bit_hamming hnsw_bit_jaccard hnsw_halfvec_cosine hnsw_halfvec_ip hnsw_halfvec_l2 hnsw_options hnsw_sparsevec_cosine hnsw_sparsevec_ip hnsw_sparsevec_l2 hnsw_unlogged hnsw_vector_cosine hnsw_vector_ip hnsw_vector_l2 ivfflat_options ivfflat_unlogged ivfflat_vector_cosine ivfflat_vector_ip ivfflat_vector_l2 sparsevec_functions sparsevec_input vector_functions vector_input
|
||||
REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
|
||||
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
|
||||
|
||||
# For /arch flags
|
||||
|
||||
279
README.md
279
README.md
@@ -20,15 +20,15 @@ Compile and install the extension (supports Postgres 12+)
|
||||
|
||||
```sh
|
||||
cd /tmp
|
||||
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
make
|
||||
make install # may need sudo
|
||||
```
|
||||
|
||||
See the [installation notes](#installation-notes---linux-and-mac) if you run into issues
|
||||
See the [installation notes](#installation-notes) if you run into issues
|
||||
|
||||
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
|
||||
You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
|
||||
|
||||
### Windows
|
||||
|
||||
@@ -44,15 +44,12 @@ Then use `nmake` to build:
|
||||
|
||||
```cmd
|
||||
set "PGROOT=C:\Program Files\PostgreSQL\16"
|
||||
cd %TEMP%
|
||||
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
nmake /F Makefile.win
|
||||
nmake /F Makefile.win install
|
||||
```
|
||||
|
||||
See the [installation notes](#installation-notes---windows) if you run into issues
|
||||
|
||||
You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
|
||||
|
||||
## Getting Started
|
||||
@@ -105,12 +102,6 @@ Insert vectors
|
||||
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
|
||||
```
|
||||
|
||||
Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py))
|
||||
|
||||
```sql
|
||||
COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
|
||||
```
|
||||
|
||||
Upsert vectors
|
||||
|
||||
```sql
|
||||
@@ -221,24 +212,7 @@ Cosine distance
|
||||
CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
|
||||
```
|
||||
|
||||
Hamming distance - unreleased
|
||||
|
||||
```sql
|
||||
CREATE INDEX ON items USING hnsw (embedding bit_hamming_ops);
|
||||
```
|
||||
|
||||
Jaccard distance - unreleased
|
||||
|
||||
```sql
|
||||
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.
|
||||
|
||||
### Index Options
|
||||
|
||||
@@ -290,8 +264,6 @@ HINT: Increase maintenance_work_mem to speed up builds.
|
||||
|
||||
Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
|
||||
|
||||
Like other index types, it’s faster to create an index after loading your initial data
|
||||
|
||||
Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
|
||||
|
||||
```sql
|
||||
@@ -343,9 +315,7 @@ 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
|
||||
|
||||
@@ -432,51 +402,13 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
|
||||
|
||||
## Performance
|
||||
|
||||
### Tuning
|
||||
|
||||
Use a tool like [PgTune](https://pgtune.leopard.in.ua/) to set initial values for Postgres server parameters. For instance, `shared_buffers` should typically be 25% of the server’s memory. You can find the config file with:
|
||||
|
||||
```sql
|
||||
SHOW config_file;
|
||||
```
|
||||
|
||||
And check individual settings with:
|
||||
|
||||
```sql
|
||||
SHOW shared_buffers;
|
||||
```
|
||||
|
||||
Be sure to restart Postgres for changes to take effect.
|
||||
|
||||
### Loading
|
||||
|
||||
Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py)).
|
||||
|
||||
```sql
|
||||
COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
|
||||
```
|
||||
|
||||
Add any indexes *after* loading the initial data for best performance.
|
||||
|
||||
### Indexing
|
||||
|
||||
See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
|
||||
|
||||
In production environments, create indexes concurrently to avoid blocking writes.
|
||||
|
||||
```sql
|
||||
CREATE INDEX CONCURRENTLY ...
|
||||
```
|
||||
|
||||
### Querying
|
||||
|
||||
Use `EXPLAIN ANALYZE` to debug performance.
|
||||
|
||||
```sql
|
||||
EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
|
||||
```
|
||||
|
||||
#### Exact Search
|
||||
### Exact Search
|
||||
|
||||
To speed up queries without an index, increase `max_parallel_workers_per_gather`.
|
||||
|
||||
@@ -490,7 +422,7 @@ If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform
|
||||
SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
|
||||
```
|
||||
|
||||
#### Approximate Search
|
||||
### Approximate Search
|
||||
|
||||
To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
|
||||
|
||||
@@ -498,7 +430,7 @@ To speed up queries with an IVFFlat index, increase the number of inverted lists
|
||||
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
|
||||
```
|
||||
|
||||
### Vacuuming
|
||||
## Vacuuming
|
||||
|
||||
Vacuuming can take a while for HNSW indexes. Speed it up by reindexing first.
|
||||
|
||||
@@ -507,41 +439,6 @@ REINDEX INDEX CONCURRENTLY index_name;
|
||||
VACUUM table_name;
|
||||
```
|
||||
|
||||
## Monitoring
|
||||
|
||||
Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
|
||||
|
||||
```sql
|
||||
CREATE EXTENSION pg_stat_statements;
|
||||
```
|
||||
|
||||
Get the most time-consuming queries with:
|
||||
|
||||
```sql
|
||||
SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
|
||||
ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
|
||||
FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
|
||||
```
|
||||
|
||||
Note: Replace `total_plan_time + total_exec_time` with `total_time` for Postgres < 13
|
||||
|
||||
Monitor recall by comparing results from approximate search with exact search.
|
||||
|
||||
```sql
|
||||
BEGIN;
|
||||
SET LOCAL enable_indexscan = off; -- use exact search
|
||||
SELECT ...
|
||||
COMMIT;
|
||||
```
|
||||
|
||||
## Scaling
|
||||
|
||||
Scale pgvector the same way you scale Postgres.
|
||||
|
||||
Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
|
||||
|
||||
Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus.py)).
|
||||
|
||||
## Languages
|
||||
|
||||
Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
|
||||
@@ -635,18 +532,6 @@ 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:
|
||||
@@ -659,17 +544,7 @@ SELECT pg_size_pretty(pg_relation_size('index_name'));
|
||||
|
||||
#### Why isn’t a query using an index?
|
||||
|
||||
The query needs to have an `ORDER BY` and `LIMIT`, and the `ORDER BY` must be the result of a distance operator, not an expression.
|
||||
|
||||
```sql
|
||||
-- index
|
||||
ORDER BY embedding <=> '[3,1,2]' LIMIT 5;
|
||||
|
||||
-- no index
|
||||
ORDER BY 1 - (embedding <=> '[3,1,2]') DESC LIMIT 5;
|
||||
```
|
||||
|
||||
You can encourage the planner to use an index for a query with:
|
||||
The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
|
||||
|
||||
```sql
|
||||
BEGIN;
|
||||
@@ -698,12 +573,6 @@ or choose to store vectors inline:
|
||||
ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
|
||||
```
|
||||
|
||||
#### Why are there less results for a query after adding an HNSW index?
|
||||
|
||||
Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
|
||||
|
||||
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
|
||||
|
||||
#### Why are there less results for a query after adding an IVFFlat index?
|
||||
|
||||
The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
|
||||
@@ -712,20 +581,11 @@ The index was likely created with too little data for the number of lists. Drop
|
||||
DROP INDEX index_name;
|
||||
```
|
||||
|
||||
Results can also be limited by the number of probes (`ivfflat.probes`).
|
||||
|
||||
Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
|
||||
|
||||
## Reference
|
||||
|
||||
- [Vector](#vector-type)
|
||||
- [Halfvec](#halfvec-type)
|
||||
- [Bit](#bit-type)
|
||||
- [Sparsevec](#sparsevec-type)
|
||||
|
||||
### Vector Type
|
||||
|
||||
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single-precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
|
||||
Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
|
||||
|
||||
### Vector Operators
|
||||
|
||||
@@ -746,81 +606,17 @@ cosine_distance(vector, vector) → double precision | cosine distance |
|
||||
inner_product(vector, vector) → double precision | inner product |
|
||||
l2_distance(vector, vector) → double precision | Euclidean distance |
|
||||
l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
|
||||
quantize_binary(vector) → bit | quantize | unreleased
|
||||
subvector(vector, integer, integer) → vector | subvector | unreleased
|
||||
vector_dims(vector) → integer | number of dimensions |
|
||||
vector_norm(vector) → double precision | Euclidean norm |
|
||||
|
||||
### Vector Aggregate Functions
|
||||
### Aggregate Functions
|
||||
|
||||
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.
|
||||
|
||||
### Bit Operators
|
||||
|
||||
Operator | Description | Added
|
||||
--- | --- | ---
|
||||
<~> | Hamming distance | unreleased
|
||||
<%> | Jaccard distance | unreleased
|
||||
|
||||
### Bit Functions
|
||||
|
||||
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 100,000 dimensions.
|
||||
|
||||
### 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
|
||||
## Installation Notes
|
||||
|
||||
### Postgres Location
|
||||
|
||||
@@ -860,26 +656,6 @@ Note: Replace `16` with your Postgres server version
|
||||
|
||||
If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
|
||||
|
||||
### Portability
|
||||
|
||||
By default, pgvector compiles with `-march=native` on some platforms for best performance. However, this can lead to `Illegal instruction` errors if trying to run the compiled extension on a different machine.
|
||||
|
||||
To compile for portability, use:
|
||||
|
||||
```sh
|
||||
make OPTFLAGS=""
|
||||
```
|
||||
|
||||
## Installation Notes - Windows
|
||||
|
||||
### Missing Header
|
||||
|
||||
If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
|
||||
|
||||
### Permissions
|
||||
|
||||
If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
|
||||
|
||||
## Additional Installation Methods
|
||||
|
||||
### Docker
|
||||
@@ -895,7 +671,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
|
||||
You can also build the image manually:
|
||||
|
||||
```sh
|
||||
git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
|
||||
git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
|
||||
cd pgvector
|
||||
docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
|
||||
```
|
||||
@@ -940,21 +716,6 @@ sudo dnf install pgvector_16
|
||||
|
||||
Note: Replace `16` with your Postgres server version
|
||||
|
||||
### pkg
|
||||
|
||||
Install the FreeBSD package with:
|
||||
|
||||
```sh
|
||||
pkg install postgresql15-pg_vector
|
||||
```
|
||||
|
||||
or the port with:
|
||||
|
||||
```sh
|
||||
cd /usr/ports/databases/pgvector
|
||||
make install
|
||||
```
|
||||
|
||||
### conda-forge
|
||||
|
||||
With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
|
||||
@@ -1064,12 +825,6 @@ make installcheck REGRESS=functions # regression test
|
||||
make prove_installcheck PROVE_TESTS=test/t/001_ivfflat_wal.pl # TAP test
|
||||
```
|
||||
|
||||
To enable assertions:
|
||||
|
||||
```sh
|
||||
make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
|
||||
```
|
||||
|
||||
To enable benchmarking:
|
||||
|
||||
```sh
|
||||
@@ -1082,6 +837,12 @@ To show memory usage:
|
||||
make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
|
||||
```
|
||||
|
||||
To enable assertions:
|
||||
|
||||
```sh
|
||||
make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
|
||||
```
|
||||
|
||||
To get k-means metrics:
|
||||
|
||||
```sh
|
||||
|
||||
@@ -1,16 +0,0 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.1'" to load this file. \quit
|
||||
|
||||
DROP OPERATOR - (vector, vector);
|
||||
|
||||
CREATE OPERATOR - (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
|
||||
);
|
||||
|
||||
ALTER OPERATOR <= (vector, vector) SET (
|
||||
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
|
||||
);
|
||||
|
||||
ALTER OPERATOR >= (vector, vector) SET (
|
||||
RESTRICT = scalargesel, JOIN = scalargejoinsel
|
||||
);
|
||||
@@ -1,2 +0,0 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.2'" to load this file. \quit
|
||||
@@ -1,255 +0,0 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.0'" to load this file. \quit
|
||||
|
||||
CREATE FUNCTION quantize_binary(vector) RETURNS bit
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION subvector(vector, int, int) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE OPERATOR <~> (
|
||||
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
|
||||
COMMUTATOR = '<~>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <%> (
|
||||
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
|
||||
COMMUTATOR = '<%>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS bit_hamming_ops
|
||||
FOR TYPE bit USING hnsw AS
|
||||
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 hamming_distance(bit, bit);
|
||||
|
||||
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_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,
|
||||
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);
|
||||
310
sql/vector.sql
310
sql/vector.sql
@@ -1,7 +1,7 @@
|
||||
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
|
||||
\echo Use "CREATE EXTENSION vector" to load this file. \quit
|
||||
|
||||
-- vector type
|
||||
-- type
|
||||
|
||||
CREATE TYPE vector;
|
||||
|
||||
@@ -29,7 +29,7 @@ CREATE TYPE vector (
|
||||
STORAGE = external
|
||||
);
|
||||
|
||||
-- vector functions
|
||||
-- functions
|
||||
|
||||
CREATE FUNCTION l2_distance(vector, vector) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
@@ -58,13 +58,7 @@ CREATE FUNCTION vector_sub(vector, vector) RETURNS vector
|
||||
CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION quantize_binary(vector) RETURNS bit
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION subvector(vector, int, int) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- vector private functions
|
||||
-- private functions
|
||||
|
||||
CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
@@ -105,7 +99,7 @@ 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;
|
||||
|
||||
-- vector aggregates
|
||||
-- aggregates
|
||||
|
||||
CREATE AGGREGATE avg(vector) (
|
||||
SFUNC = vector_accum,
|
||||
@@ -123,7 +117,7 @@ CREATE AGGREGATE sum(vector) (
|
||||
PARALLEL = SAFE
|
||||
);
|
||||
|
||||
-- vector cast functions
|
||||
-- cast functions
|
||||
|
||||
CREATE FUNCTION vector(vector, integer, boolean) RETURNS vector
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
@@ -143,7 +137,7 @@ CREATE FUNCTION array_to_vector(numeric[], integer, boolean) RETURNS vector
|
||||
CREATE FUNCTION vector_to_float4(vector, integer, boolean) RETURNS real[]
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
-- vector casts
|
||||
-- casts
|
||||
|
||||
CREATE CAST (vector AS vector)
|
||||
WITH FUNCTION vector(vector, integer, boolean) AS IMPLICIT;
|
||||
@@ -163,7 +157,7 @@ CREATE CAST (double precision[] AS vector)
|
||||
CREATE CAST (numeric[] AS vector)
|
||||
WITH FUNCTION array_to_vector(numeric[], integer, boolean) AS ASSIGNMENT;
|
||||
|
||||
-- vector operators
|
||||
-- operators
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = l2_distance,
|
||||
@@ -186,7 +180,8 @@ CREATE OPERATOR + (
|
||||
);
|
||||
|
||||
CREATE OPERATOR - (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub,
|
||||
COMMUTATOR = -
|
||||
);
|
||||
|
||||
CREATE OPERATOR * (
|
||||
@@ -200,10 +195,11 @@ CREATE OPERATOR < (
|
||||
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
|
||||
);
|
||||
|
||||
-- should use scalarlesel and scalarlejoinsel, but not supported in Postgres < 11
|
||||
CREATE OPERATOR <= (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_le,
|
||||
COMMUTATOR = >= , NEGATOR = > ,
|
||||
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
|
||||
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR = (
|
||||
@@ -218,10 +214,11 @@ CREATE OPERATOR <> (
|
||||
RESTRICT = eqsel, JOIN = eqjoinsel
|
||||
);
|
||||
|
||||
-- should use scalargesel and scalargejoinsel, but not supported in Postgres < 11
|
||||
CREATE OPERATOR >= (
|
||||
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_ge,
|
||||
COMMUTATOR = <= , NEGATOR = < ,
|
||||
RESTRICT = scalargesel, JOIN = scalargejoinsel
|
||||
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
|
||||
);
|
||||
|
||||
CREATE OPERATOR > (
|
||||
@@ -246,7 +243,7 @@ CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
|
||||
|
||||
COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
|
||||
|
||||
-- vector opclasses
|
||||
-- opclasses
|
||||
|
||||
CREATE OPERATOR CLASS vector_ops
|
||||
DEFAULT FOR TYPE vector USING btree AS
|
||||
@@ -293,282 +290,3 @@ CREATE OPERATOR CLASS vector_cosine_ops
|
||||
OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 vector_negative_inner_product(vector, vector),
|
||||
FUNCTION 2 vector_norm(vector);
|
||||
|
||||
-- bit functions
|
||||
|
||||
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
|
||||
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE OPERATOR <~> (
|
||||
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
|
||||
COMMUTATOR = '<~>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR <%> (
|
||||
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
|
||||
COMMUTATOR = '<%>'
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS bit_hamming_ops
|
||||
FOR TYPE bit USING hnsw AS
|
||||
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 hamming_distance(bit, bit);
|
||||
|
||||
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_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,
|
||||
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);
|
||||
|
||||
@@ -1,90 +0,0 @@
|
||||
#include "postgres.h"
|
||||
|
||||
#include "bitvector.h"
|
||||
#include "port/pg_bitutils.h"
|
||||
#include "utils/varbit.h"
|
||||
|
||||
#if PG_VERSION_NUM >= 160000
|
||||
#include "varatt.h"
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Allocate and initialize a new bit vector
|
||||
*/
|
||||
VarBit *
|
||||
InitBitVector(int dim)
|
||||
{
|
||||
VarBit *result;
|
||||
int size;
|
||||
|
||||
size = VARBITTOTALLEN(dim);
|
||||
result = (VarBit *) palloc0(size);
|
||||
SET_VARSIZE(result, size);
|
||||
VARBITLEN(result) = dim;
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure same dimensions
|
||||
*/
|
||||
static inline void
|
||||
CheckDims(VarBit *a, VarBit *b)
|
||||
{
|
||||
if (VARBITLEN(a) != VARBITLEN(b))
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("different bit lengths %u and %u", VARBITLEN(a), VARBITLEN(b))));
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the Hamming distance between two bit vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(hamming_distance);
|
||||
Datum
|
||||
hamming_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
VarBit *a = PG_GETARG_VARBIT_P(0);
|
||||
VarBit *b = PG_GETARG_VARBIT_P(1);
|
||||
unsigned char *ax = VARBITS(a);
|
||||
unsigned char *bx = VARBITS(b);
|
||||
uint64 distance = 0;
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
/* TODO Improve performance */
|
||||
for (uint32 i = 0; i < VARBITBYTES(a); i++)
|
||||
distance += pg_number_of_ones[ax[i] ^ bx[i]];
|
||||
|
||||
PG_RETURN_FLOAT8((double) distance);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get the Jaccard distance between two bit vectors
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(jaccard_distance);
|
||||
Datum
|
||||
jaccard_distance(PG_FUNCTION_ARGS)
|
||||
{
|
||||
VarBit *a = PG_GETARG_VARBIT_P(0);
|
||||
VarBit *b = PG_GETARG_VARBIT_P(1);
|
||||
unsigned char *ax = VARBITS(a);
|
||||
unsigned char *bx = VARBITS(b);
|
||||
uint64 ab = 0;
|
||||
uint64 aa;
|
||||
uint64 bb;
|
||||
|
||||
CheckDims(a, b);
|
||||
|
||||
/* TODO Improve performance */
|
||||
for (uint32 i = 0; i < VARBITBYTES(a); i++)
|
||||
ab += pg_number_of_ones[ax[i] & bx[i]];
|
||||
|
||||
if (ab == 0)
|
||||
PG_RETURN_FLOAT8(1);
|
||||
|
||||
aa = pg_popcount((char *) ax, VARBITBYTES(a));
|
||||
bb = pg_popcount((char *) bx, VARBITBYTES(b));
|
||||
|
||||
PG_RETURN_FLOAT8(1 - (ab / ((double) (aa + bb - ab))));
|
||||
}
|
||||
@@ -1,8 +0,0 @@
|
||||
#ifndef BITVECTOR_H
|
||||
#define BITVECTOR_H
|
||||
|
||||
#include "utils/varbit.h"
|
||||
|
||||
VarBit *InitBitVector(int dim);
|
||||
|
||||
#endif
|
||||
1020
src/halfvec.c
1020
src/halfvec.c
File diff suppressed because it is too large
Load Diff
@@ -1,43 +0,0 @@
|
||||
#ifndef HALFVEC_H
|
||||
#define HALFVEC_H
|
||||
|
||||
#define __STDC_WANT_IEC_60559_TYPES_EXT__
|
||||
|
||||
#include <float.h>
|
||||
|
||||
#include "vector.h"
|
||||
|
||||
#ifdef __FLT16_MAX__
|
||||
#define FLT16_SUPPORT
|
||||
#endif
|
||||
|
||||
#ifdef FLT16_SUPPORT
|
||||
#define half _Float16
|
||||
#define HALF_MAX FLT16_MAX
|
||||
#else
|
||||
/* TODO #pragma message("")? */
|
||||
#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
|
||||
@@ -8,7 +8,6 @@
|
||||
#include "commands/progress.h"
|
||||
#include "commands/vacuum.h"
|
||||
#include "hnsw.h"
|
||||
#include "miscadmin.h"
|
||||
#include "utils/guc.h"
|
||||
#include "utils/selfuncs.h"
|
||||
|
||||
@@ -29,7 +28,7 @@ static relopt_kind hnsw_relopt_kind;
|
||||
* this grows bigger, we should use a shmem_request_hook and
|
||||
* RequestAddinShmemSpace() to pre-reserve space for this.
|
||||
*/
|
||||
void
|
||||
static void
|
||||
HnswInitLockTranche(void)
|
||||
{
|
||||
int *tranche_ids;
|
||||
@@ -54,8 +53,7 @@ HnswInitLockTranche(void)
|
||||
void
|
||||
HnswInit(void)
|
||||
{
|
||||
if (!process_shared_preload_libraries_in_progress)
|
||||
HnswInitLockTranche();
|
||||
HnswInitLockTranche();
|
||||
|
||||
hnsw_relopt_kind = add_reloption_kind();
|
||||
add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",
|
||||
|
||||
27
src/hnsw.h
27
src/hnsw.h
@@ -17,7 +17,6 @@
|
||||
#endif
|
||||
|
||||
#define HNSW_MAX_DIM 2000
|
||||
#define HNSW_MAX_NNZ 1000
|
||||
|
||||
/* Support functions */
|
||||
#define HNSW_DISTANCE_PROC 1
|
||||
@@ -56,14 +55,6 @@
|
||||
#define HNSW_UPDATE_ENTRY_GREATER 1
|
||||
#define HNSW_UPDATE_ENTRY_ALWAYS 2
|
||||
|
||||
typedef enum HnswType
|
||||
{
|
||||
HNSW_TYPE_VECTOR,
|
||||
HNSW_TYPE_HALFVEC,
|
||||
HNSW_TYPE_BIT,
|
||||
HNSW_TYPE_SPARSEVEC
|
||||
} HnswType;
|
||||
|
||||
/* Build phases */
|
||||
/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
|
||||
#define PROGRESS_HNSW_PHASE_LOAD 2
|
||||
@@ -89,7 +80,7 @@ typedef enum HnswType
|
||||
|
||||
#if PG_VERSION_NUM < 130000
|
||||
#define list_delete_last(list) list_truncate(list, list_length(list) - 1)
|
||||
#define list_sort(list, cmp) ((list) = list_qsort(list, cmp))
|
||||
#define list_sort(list, cmp) list_qsort(list, cmp)
|
||||
#endif
|
||||
|
||||
#define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
|
||||
@@ -138,7 +129,7 @@ HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr)
|
||||
HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
|
||||
HnswPtrDeclare(char, DatumRelptr, DatumPtr);
|
||||
|
||||
struct HnswElementData
|
||||
typedef struct HnswElementData
|
||||
{
|
||||
HnswElementPtr next;
|
||||
ItemPointerData heaptids[HNSW_HEAPTIDS];
|
||||
@@ -153,7 +144,7 @@ struct HnswElementData
|
||||
BlockNumber neighborPage;
|
||||
DatumPtr value;
|
||||
LWLock lock;
|
||||
};
|
||||
} HnswElementData;
|
||||
|
||||
typedef HnswElementData * HnswElement;
|
||||
|
||||
@@ -164,12 +155,12 @@ typedef struct HnswCandidate
|
||||
bool closer;
|
||||
} HnswCandidate;
|
||||
|
||||
struct HnswNeighborArray
|
||||
typedef struct HnswNeighborArray
|
||||
{
|
||||
int length;
|
||||
bool closerSet;
|
||||
HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
|
||||
};
|
||||
} HnswNeighborArray;
|
||||
|
||||
typedef struct HnswPairingHeapNode
|
||||
{
|
||||
@@ -194,7 +185,6 @@ typedef struct HnswGraph
|
||||
|
||||
/* Entry state */
|
||||
LWLock entryLock;
|
||||
LWLock entryWaitLock;
|
||||
HnswElementPtr entryPoint;
|
||||
|
||||
/* Allocations state */
|
||||
@@ -251,7 +241,6 @@ typedef struct HnswBuildState
|
||||
Relation index;
|
||||
IndexInfo *indexInfo;
|
||||
ForkNumber forkNum;
|
||||
HnswType type;
|
||||
|
||||
/* Settings */
|
||||
int dimensions;
|
||||
@@ -272,6 +261,7 @@ typedef struct HnswBuildState
|
||||
HnswGraph *graph;
|
||||
double ml;
|
||||
int maxLevel;
|
||||
Vector *normvec;
|
||||
|
||||
/* Memory */
|
||||
MemoryContext graphCtx;
|
||||
@@ -376,9 +366,7 @@ typedef struct HnswVacuumState
|
||||
int HnswGetM(Relation index);
|
||||
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);
|
||||
bool HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
|
||||
Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
|
||||
void HnswInitPage(Buffer buf, Page page);
|
||||
void HnswInit(void);
|
||||
@@ -401,7 +389,6 @@ void HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation i
|
||||
void HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
|
||||
void HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
|
||||
void HnswLoadNeighbors(HnswElement element, Relation index, int m);
|
||||
void HnswInitLockTranche(void);
|
||||
PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
|
||||
|
||||
/* Index access methods */
|
||||
|
||||
@@ -44,7 +44,6 @@
|
||||
#include "access/xact.h"
|
||||
#include "access/xloginsert.h"
|
||||
#include "catalog/index.h"
|
||||
#include "catalog/pg_type_d.h"
|
||||
#include "commands/progress.h"
|
||||
#include "hnsw.h"
|
||||
#include "miscadmin.h"
|
||||
@@ -177,7 +176,7 @@ CreateGraphPages(HnswBuildState * buildstate)
|
||||
Size etupSize;
|
||||
Size ntupSize;
|
||||
Size combinedSize;
|
||||
Pointer valuePtr = HnswPtrAccess(base, element->value);
|
||||
void *valuePtr = HnswPtrAccess(base, element->value);
|
||||
|
||||
/* Update iterator */
|
||||
iter = element->next;
|
||||
@@ -432,15 +431,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
|
||||
HnswGraph *graph = buildstate->graph;
|
||||
HnswElement entryPoint;
|
||||
LWLock *entryLock = &graph->entryLock;
|
||||
LWLock *entryWaitLock = &graph->entryWaitLock;
|
||||
int efConstruction = buildstate->efConstruction;
|
||||
int m = buildstate->m;
|
||||
char *base = buildstate->hnswarea;
|
||||
|
||||
/* Wait if another process needs exclusive lock on entry lock */
|
||||
LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
|
||||
LWLockRelease(entryWaitLock);
|
||||
|
||||
/* Get entry point */
|
||||
LWLockAcquire(entryLock, LW_SHARED);
|
||||
entryPoint = HnswPtrAccess(base, graph->entryPoint);
|
||||
@@ -451,10 +445,8 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
|
||||
/* Release shared lock */
|
||||
LWLockRelease(entryLock);
|
||||
|
||||
/* Tell other processes to wait and get exclusive lock */
|
||||
LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
|
||||
/* Get exclusive lock */
|
||||
LWLockAcquire(entryLock, LW_EXCLUSIVE);
|
||||
LWLockRelease(entryWaitLock);
|
||||
|
||||
/* Get latest entry point after lock is acquired */
|
||||
entryPoint = HnswPtrAccess(base, graph->entryPoint);
|
||||
@@ -487,13 +479,10 @@ 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)
|
||||
{
|
||||
if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
|
||||
if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
|
||||
return false;
|
||||
}
|
||||
|
||||
@@ -612,9 +601,6 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
|
||||
static void
|
||||
InitGraph(HnswGraph * graph, char *base, long memoryTotal)
|
||||
{
|
||||
/* Initialize the lock tranche if needed */
|
||||
HnswInitLockTranche();
|
||||
|
||||
HnswPtrStore(base, graph->head, (HnswElement) NULL);
|
||||
HnswPtrStore(base, graph->entryPoint, (HnswElement) NULL);
|
||||
graph->memoryUsed = 0;
|
||||
@@ -623,7 +609,6 @@ InitGraph(HnswGraph * graph, char *base, long memoryTotal)
|
||||
graph->indtuples = 0;
|
||||
SpinLockInit(&graph->lock);
|
||||
LWLockInitialize(&graph->entryLock, hnsw_lock_tranche_id);
|
||||
LWLockInitialize(&graph->entryWaitLock, hnsw_lock_tranche_id);
|
||||
LWLockInitialize(&graph->allocatorLock, hnsw_lock_tranche_id);
|
||||
LWLockInitialize(&graph->flushLock, hnsw_lock_tranche_id);
|
||||
}
|
||||
@@ -669,22 +654,6 @@ HnswSharedMemoryAlloc(Size size, void *state)
|
||||
return chunk;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get max dimensions
|
||||
*/
|
||||
static int
|
||||
GetMaxDimensions(HnswType type)
|
||||
{
|
||||
int maxDimensions = HNSW_MAX_DIM;
|
||||
|
||||
if (type == HNSW_TYPE_HALFVEC)
|
||||
maxDimensions *= 2;
|
||||
else if (type == HNSW_TYPE_BIT)
|
||||
maxDimensions *= 32;
|
||||
|
||||
return maxDimensions;
|
||||
}
|
||||
|
||||
/*
|
||||
* Initialize the build state
|
||||
*/
|
||||
@@ -695,23 +664,13 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
|
||||
buildstate->index = index;
|
||||
buildstate->indexInfo = indexInfo;
|
||||
buildstate->forkNum = forkNum;
|
||||
buildstate->type = HnswGetType(index);
|
||||
|
||||
buildstate->m = HnswGetM(index);
|
||||
buildstate->efConstruction = HnswGetEfConstruction(index);
|
||||
buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
|
||||
|
||||
|
||||
if (buildstate->type != HNSW_TYPE_SPARSEVEC)
|
||||
{
|
||||
int maxDimensions = GetMaxDimensions(buildstate->type);
|
||||
|
||||
if (buildstate->dimensions < 0)
|
||||
elog(ERROR, "column does not have dimensions");
|
||||
|
||||
if (buildstate->dimensions > maxDimensions)
|
||||
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
|
||||
}
|
||||
if (buildstate->dimensions > HNSW_MAX_DIM)
|
||||
elog(ERROR, "column cannot have more than %d dimensions for hnsw index", HNSW_MAX_DIM);
|
||||
|
||||
if (buildstate->efConstruction < 2 * buildstate->m)
|
||||
elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
|
||||
@@ -729,6 +688,9 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
|
||||
buildstate->ml = HnswGetMl(buildstate->m);
|
||||
buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
|
||||
|
||||
/* Reuse for each tuple */
|
||||
buildstate->normvec = InitVector(buildstate->dimensions);
|
||||
|
||||
buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
|
||||
"Hnsw build graph context",
|
||||
#if PG_VERSION_NUM >= 150000
|
||||
@@ -752,6 +714,7 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
|
||||
static void
|
||||
FreeBuildState(HnswBuildState * buildstate)
|
||||
{
|
||||
pfree(buildstate->normvec);
|
||||
MemoryContextDelete(buildstate->graphCtx);
|
||||
MemoryContextDelete(buildstate->tmpCtx);
|
||||
}
|
||||
@@ -1007,14 +970,6 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
|
||||
/* Report less than allocated so never fails */
|
||||
InitGraph(&hnswshared->graphData, hnswarea, esthnswarea - 1024 * 1024);
|
||||
|
||||
/*
|
||||
* Avoid base address for relptr for Postgres < 14.5
|
||||
* https://github.com/postgres/postgres/commit/7201cd18627afc64850537806da7f22150d1a83b
|
||||
*/
|
||||
#if PG_VERSION_NUM < 140005
|
||||
hnswshared->graphData.memoryUsed += MAXALIGN(1);
|
||||
#endif
|
||||
|
||||
shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_SHARED, hnswshared);
|
||||
shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_AREA, hnswarea);
|
||||
|
||||
|
||||
@@ -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, NULL))
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
@@ -40,6 +40,29 @@ GetScanItems(IndexScanDesc scan, Datum q)
|
||||
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get dimensions from metapage
|
||||
*/
|
||||
static int
|
||||
GetDimensions(Relation index)
|
||||
{
|
||||
Buffer buf;
|
||||
Page page;
|
||||
HnswMetaPage metap;
|
||||
int dimensions;
|
||||
|
||||
buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
|
||||
LockBuffer(buf, BUFFER_LOCK_SHARE);
|
||||
page = BufferGetPage(buf);
|
||||
metap = HnswPageGetMeta(page);
|
||||
|
||||
dimensions = metap->dimensions;
|
||||
|
||||
UnlockReleaseBuffer(buf);
|
||||
|
||||
return dimensions;
|
||||
}
|
||||
|
||||
/*
|
||||
* Get scan value
|
||||
*/
|
||||
@@ -50,7 +73,7 @@ GetScanValue(IndexScanDesc scan)
|
||||
Datum value;
|
||||
|
||||
if (scan->orderByData->sk_flags & SK_ISNULL)
|
||||
value = PointerGetDatum(NULL);
|
||||
value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
|
||||
else
|
||||
{
|
||||
value = scan->orderByData->sk_argument;
|
||||
@@ -61,7 +84,7 @@ GetScanValue(IndexScanDesc scan)
|
||||
|
||||
/* Fine if normalization fails */
|
||||
if (so->normprocinfo != NULL)
|
||||
HnswNormValue(so->normprocinfo, so->collation, &value, HnswGetType(scan->indexRelation));
|
||||
HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
|
||||
}
|
||||
|
||||
return value;
|
||||
|
||||
140
src/hnswutils.c
140
src/hnswutils.c
@@ -3,17 +3,12 @@
|
||||
#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
|
||||
@@ -154,39 +149,6 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
|
||||
return index_getprocinfo(index, 1, procnum);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get type
|
||||
*/
|
||||
HnswType
|
||||
HnswGetType(Relation index)
|
||||
{
|
||||
Oid typid = TupleDescAttr(index->rd_att, 0)->atttypid;
|
||||
HeapTuple tuple;
|
||||
Form_pg_type type;
|
||||
int result;
|
||||
|
||||
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
|
||||
elog(ERROR, "Unsupported type");
|
||||
|
||||
ReleaseSysCache(tuple);
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
/*
|
||||
* Divide by the norm
|
||||
*
|
||||
@@ -196,53 +158,21 @@ HnswGetType(Relation index)
|
||||
* if it's different than the original value
|
||||
*/
|
||||
bool
|
||||
HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
|
||||
HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
|
||||
{
|
||||
double norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
|
||||
|
||||
if (norm > 0)
|
||||
{
|
||||
/* TODO Remove vector-specific code */
|
||||
if (type == HNSW_TYPE_VECTOR)
|
||||
{
|
||||
Vector *v = DatumGetVector(*value);
|
||||
Vector *result = InitVector(v->dim);
|
||||
Vector *v = DatumGetVector(*value);
|
||||
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
if (result == NULL)
|
||||
result = InitVector(v->dim);
|
||||
|
||||
*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++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
|
||||
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->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");
|
||||
*value = PointerGetDatum(result);
|
||||
|
||||
return true;
|
||||
}
|
||||
@@ -250,21 +180,6 @@ 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
|
||||
*/
|
||||
@@ -660,12 +575,7 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
|
||||
|
||||
/* Calculate distance */
|
||||
if (distance != NULL)
|
||||
{
|
||||
if (DatumGetPointer(*q) == NULL)
|
||||
*distance = 0;
|
||||
else
|
||||
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
|
||||
}
|
||||
*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
|
||||
|
||||
UnlockReleaseBuffer(buf);
|
||||
}
|
||||
@@ -950,15 +860,12 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
|
||||
static int
|
||||
#if PG_VERSION_NUM >= 130000
|
||||
CompareCandidateDistances(const ListCell *a, const ListCell *b)
|
||||
{
|
||||
HnswCandidate *hca = lfirst(a);
|
||||
HnswCandidate *hcb = lfirst(b);
|
||||
#else
|
||||
CompareCandidateDistances(const void *a, const void *b)
|
||||
{
|
||||
HnswCandidate *hca = lfirst(*(ListCell **) a);
|
||||
HnswCandidate *hcb = lfirst(*(ListCell **) b);
|
||||
#endif
|
||||
{
|
||||
HnswCandidate *hca = lfirst((ListCell *) a);
|
||||
HnswCandidate *hcb = lfirst((ListCell *) b);
|
||||
|
||||
if (hca->distance < hcb->distance)
|
||||
return 1;
|
||||
@@ -981,15 +888,12 @@ CompareCandidateDistances(const void *a, const void *b)
|
||||
static int
|
||||
#if PG_VERSION_NUM >= 130000
|
||||
CompareCandidateDistancesOffset(const ListCell *a, const ListCell *b)
|
||||
{
|
||||
HnswCandidate *hca = lfirst(a);
|
||||
HnswCandidate *hcb = lfirst(b);
|
||||
#else
|
||||
CompareCandidateDistancesOffset(const void *a, const void *b)
|
||||
{
|
||||
HnswCandidate *hca = lfirst(*(ListCell **) a);
|
||||
HnswCandidate *hcb = lfirst(*(ListCell **) b);
|
||||
#endif
|
||||
{
|
||||
HnswCandidate *hca = lfirst((ListCell *) a);
|
||||
HnswCandidate *hcb = lfirst((ListCell *) b);
|
||||
|
||||
if (hca->distance < hcb->distance)
|
||||
return 1;
|
||||
@@ -1048,9 +952,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
|
||||
{
|
||||
List *r = NIL;
|
||||
List *w = list_copy(c);
|
||||
HnswCandidate **wd;
|
||||
int wdlen = 0;
|
||||
int wdoff = 0;
|
||||
pairingheap *wd;
|
||||
HnswNeighborArray *neighbors = HnswGetNeighbors(base, e2, lc);
|
||||
bool mustCalculate = !neighbors->closerSet;
|
||||
List *added = NIL;
|
||||
@@ -1059,7 +961,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
|
||||
if (list_length(w) <= lm)
|
||||
return w;
|
||||
|
||||
wd = palloc(sizeof(HnswCandidate *) * list_length(w));
|
||||
wd = pairingheap_allocate(CompareNearestCandidates, NULL);
|
||||
|
||||
/* Ensure order of candidates is deterministic for closer caching */
|
||||
if (sortCandidates)
|
||||
@@ -1125,21 +1027,21 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
|
||||
if (e->closer)
|
||||
r = lappend(r, e);
|
||||
else
|
||||
wd[wdlen++] = e;
|
||||
pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
|
||||
}
|
||||
|
||||
/* Cached value can only be used in future if sorted deterministically */
|
||||
neighbors->closerSet = sortCandidates;
|
||||
|
||||
/* Keep pruned connections */
|
||||
while (wdoff < wdlen && list_length(r) < lm)
|
||||
r = lappend(r, wd[wdoff++]);
|
||||
while (!pairingheap_is_empty(wd) && list_length(r) < lm)
|
||||
r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
|
||||
|
||||
/* Return pruned for update connections */
|
||||
if (pruned != NULL)
|
||||
{
|
||||
if (wdoff < wdlen)
|
||||
*pruned = wd[wdoff];
|
||||
if (!pairingheap_is_empty(wd))
|
||||
*pruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner;
|
||||
else
|
||||
*pruned = linitial(w);
|
||||
}
|
||||
|
||||
@@ -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, buildstate->normvec))
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -105,7 +105,7 @@ SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
|
||||
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
|
||||
|
||||
/* Add sample */
|
||||
AddSample(values, buildstate);
|
||||
AddSample(values, state);
|
||||
|
||||
/* Reset memory context */
|
||||
MemoryContextSwitchTo(oldCtx);
|
||||
@@ -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, buildstate->normvec))
|
||||
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;
|
||||
@@ -370,6 +356,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
|
||||
buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions);
|
||||
buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
|
||||
|
||||
/* Reuse for each tuple */
|
||||
buildstate->normvec = InitVector(buildstate->dimensions);
|
||||
|
||||
buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
|
||||
"Ivfflat build temporary context",
|
||||
ALLOCSET_DEFAULT_SIZES);
|
||||
@@ -391,6 +380,7 @@ FreeBuildState(IvfflatBuildState * buildstate)
|
||||
{
|
||||
VectorArrayFree(buildstate->centers);
|
||||
pfree(buildstate->listInfo);
|
||||
pfree(buildstate->normvec);
|
||||
|
||||
#ifdef IVFFLAT_KMEANS_DEBUG
|
||||
pfree(buildstate->listSums);
|
||||
|
||||
@@ -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;
|
||||
@@ -178,6 +172,7 @@ typedef struct IvfflatBuildState
|
||||
VectorArray samples;
|
||||
VectorArray centers;
|
||||
ListInfo *listInfo;
|
||||
Vector *normvec;
|
||||
|
||||
#ifdef IVFFLAT_KMEANS_DEBUG
|
||||
double inertia;
|
||||
@@ -272,8 +267,7 @@ 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, Vector * result);
|
||||
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, NULL))
|
||||
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, NULL);
|
||||
}
|
||||
|
||||
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 +75,21 @@ 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, Vector * result)
|
||||
{
|
||||
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);
|
||||
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
if (result == NULL)
|
||||
result = InitVector(v->dim);
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
}
|
||||
else
|
||||
elog(ERROR, "Unsupported type");
|
||||
for (int i = 0; i < v->dim; i++)
|
||||
result->x[i] = v->x[i] / norm;
|
||||
|
||||
*value = PointerGetDatum(result);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
658
src/sparsevec.c
658
src/sparsevec.c
@@ -1,658 +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 valid nnz
|
||||
*/
|
||||
static inline void
|
||||
CheckNnz(int nnz)
|
||||
{
|
||||
if (nnz < 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("sparsevec must have at least one element")));
|
||||
|
||||
if (nnz > SPARSEVEC_MAX_NNZ)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
|
||||
errmsg("sparsevec cannot have more elements than non-zero elements")));
|
||||
}
|
||||
|
||||
/*
|
||||
* Ensure valid index
|
||||
*/
|
||||
static inline void
|
||||
CheckIndex(int32 *indices, int i)
|
||||
{
|
||||
int32 index = indices[i];
|
||||
|
||||
if (index < 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("index must not be negative")));
|
||||
|
||||
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 nnz)
|
||||
{
|
||||
SparseVector *result;
|
||||
int size;
|
||||
|
||||
size = SPARSEVEC_SIZE(nnz);
|
||||
result = (SparseVector *) palloc0(size);
|
||||
SET_VARSIZE(result, size);
|
||||
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);
|
||||
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 < 0 || 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++;
|
||||
|
||||
/* 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 right brace.")));
|
||||
|
||||
pfree(litcopy);
|
||||
|
||||
result = InitSparseVector(nnz);
|
||||
rvalues = SPARSEVEC_VALUES(result);
|
||||
for (int i = 0; i < nnz; i++)
|
||||
{
|
||||
result->indices[i] = indices[i];
|
||||
rvalues[i] = values[i];
|
||||
|
||||
CheckIndex(result->indices, i);
|
||||
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 ,
|
||||
*
|
||||
* 3 bytes for {, }, and \0
|
||||
*/
|
||||
buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 2);
|
||||
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, '}');
|
||||
*ptr = '\0';
|
||||
|
||||
PG_FREE_IF_COPY(sparsevec, 0);
|
||||
PG_RETURN_CSTRING(buf);
|
||||
}
|
||||
|
||||
/*
|
||||
* 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);
|
||||
SparseVector *result;
|
||||
int32 nnz;
|
||||
int32 unused;
|
||||
int32 unused2;
|
||||
float *values;
|
||||
|
||||
nnz = pq_getmsgint(buf, sizeof(int32));
|
||||
unused = pq_getmsgint(buf, sizeof(int32));
|
||||
unused2 = pq_getmsgint(buf, sizeof(int32));
|
||||
|
||||
CheckNnz(nnz);
|
||||
|
||||
if (unused != 0 || unused2 != 0)
|
||||
ereport(ERROR,
|
||||
(errcode(ERRCODE_DATA_EXCEPTION),
|
||||
errmsg("expected unused to be 0, not %d", unused)));
|
||||
|
||||
result = InitSparseVector(nnz);
|
||||
values = SPARSEVEC_VALUES(result);
|
||||
|
||||
for (int i = 0; i < nnz; i++)
|
||||
{
|
||||
result->indices[i] = pq_getmsgint(buf, sizeof(int32));
|
||||
CheckIndex(result->indices, i);
|
||||
}
|
||||
|
||||
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->nnz, sizeof(int32));
|
||||
pq_sendint(&buf, svec->unused, sizeof(int32));
|
||||
pq_sendint(&buf, svec->unused2, sizeof(int32));
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
pq_sendint(&buf, svec->indices[i], sizeof(int32));
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
pq_sendfloat4(&buf, values[i]);
|
||||
|
||||
PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
|
||||
}
|
||||
|
||||
/*
|
||||
* Convert sparse vector to sparse vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec);
|
||||
Datum
|
||||
sparsevec(PG_FUNCTION_ARGS)
|
||||
{
|
||||
SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
|
||||
|
||||
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);
|
||||
SparseVector *result;
|
||||
int dim = vec->dim;
|
||||
int nnz = 0;
|
||||
float *values;
|
||||
int j = 0;
|
||||
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
if (vec->x[i] != 0)
|
||||
nnz++;
|
||||
}
|
||||
|
||||
CheckNnz(nnz);
|
||||
result = InitSparseVector(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;
|
||||
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);
|
||||
|
||||
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);
|
||||
|
||||
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);
|
||||
|
||||
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);
|
||||
|
||||
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;
|
||||
|
||||
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));
|
||||
}
|
||||
@@ -1,24 +0,0 @@
|
||||
#ifndef SPARSEVEC_H
|
||||
#define SPARSEVEC_H
|
||||
|
||||
#define SPARSEVEC_MAX_NNZ 100000
|
||||
|
||||
/* 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 nnz;
|
||||
int32 unused;
|
||||
int32 unused2;
|
||||
int32 indices[FLEXIBLE_ARRAY_MEMBER];
|
||||
} SparseVector;
|
||||
|
||||
SparseVector *InitSparseVector(int nnz);
|
||||
|
||||
#endif
|
||||
160
src/vector.c
160
src/vector.c
@@ -2,17 +2,14 @@
|
||||
|
||||
#include <math.h>
|
||||
|
||||
#include "bitvector.h"
|
||||
#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 +197,6 @@ 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 +212,15 @@ 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++;
|
||||
|
||||
@@ -288,9 +275,6 @@ vector_in(PG_FUNCTION_ARGS)
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
#define AppendChar(ptr, c) (*(ptr)++ = (c))
|
||||
#define AppendFloat(ptr, f) ((ptr) += float_to_shortest_decimal_bufn((f), (ptr)))
|
||||
|
||||
/*
|
||||
* Convert internal representation to textual representation
|
||||
*/
|
||||
@@ -302,6 +286,7 @@ vector_out(PG_FUNCTION_ARGS)
|
||||
int dim = vector->dim;
|
||||
char *buf;
|
||||
char *ptr;
|
||||
int n;
|
||||
|
||||
/*
|
||||
* Need:
|
||||
@@ -316,17 +301,21 @@ vector_out(PG_FUNCTION_ARGS)
|
||||
buf = (char *) palloc(FLOAT_SHORTEST_DECIMAL_LEN * dim + 2);
|
||||
ptr = buf;
|
||||
|
||||
AppendChar(ptr, '[');
|
||||
|
||||
*ptr = '[';
|
||||
ptr++;
|
||||
for (int i = 0; i < dim; i++)
|
||||
{
|
||||
if (i > 0)
|
||||
AppendChar(ptr, ',');
|
||||
{
|
||||
*ptr = ',';
|
||||
ptr++;
|
||||
}
|
||||
|
||||
AppendFloat(ptr, vector->x[i]);
|
||||
n = float_to_shortest_decimal_bufn(vector->x[i], ptr);
|
||||
ptr += n;
|
||||
}
|
||||
|
||||
AppendChar(ptr, ']');
|
||||
*ptr = ']';
|
||||
ptr++;
|
||||
*ptr = '\0';
|
||||
|
||||
PG_FREE_IF_COPY(vector, 0);
|
||||
@@ -543,28 +532,6 @@ 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
|
||||
*/
|
||||
@@ -893,56 +860,6 @@ vector_mul(PG_FUNCTION_ARGS)
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Quantize a vector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(quantize_binary);
|
||||
Datum
|
||||
quantize_binary(PG_FUNCTION_ARGS)
|
||||
{
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
float *ax = a->x;
|
||||
VarBit *result = InitBitVector(a->dim);
|
||||
unsigned char *rx = VARBITS(result);
|
||||
|
||||
for (int i = 0; i < a->dim; i++)
|
||||
rx[i / 8] |= (ax[i] > 0) << (7 - (i % 8));
|
||||
|
||||
PG_RETURN_VARBIT_P(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Get a subvector
|
||||
*/
|
||||
PGDLLEXPORT PG_FUNCTION_INFO_V1(subvector);
|
||||
Datum
|
||||
subvector(PG_FUNCTION_ARGS)
|
||||
{
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
int32 start = PG_GETARG_INT32(1);
|
||||
int32 count = PG_GETARG_INT32(2);
|
||||
int32 end = start + count;
|
||||
float *ax = a->x;
|
||||
Vector *result;
|
||||
int dim;
|
||||
|
||||
/* Indexing starts at 1, like substring */
|
||||
if (start < 1)
|
||||
start = 1;
|
||||
|
||||
if (end > a->dim)
|
||||
end = a->dim + 1;
|
||||
|
||||
dim = end - start;
|
||||
CheckDim(dim);
|
||||
result = InitVector(dim);
|
||||
|
||||
for (int i = 0; i < dim; i++)
|
||||
result->x[i] = ax[start - 1 + i];
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
/*
|
||||
* Internal helper to compare vectors
|
||||
*/
|
||||
@@ -980,6 +897,9 @@ vector_lt(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) < 0);
|
||||
}
|
||||
|
||||
@@ -993,6 +913,9 @@ vector_le(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) <= 0);
|
||||
}
|
||||
|
||||
@@ -1006,6 +929,9 @@ vector_eq(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) == 0);
|
||||
}
|
||||
|
||||
@@ -1019,6 +945,9 @@ vector_ne(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) != 0);
|
||||
}
|
||||
|
||||
@@ -1032,6 +961,9 @@ vector_ge(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) >= 0);
|
||||
}
|
||||
|
||||
@@ -1045,6 +977,9 @@ vector_gt(PG_FUNCTION_ARGS)
|
||||
Vector *a = PG_GETARG_VECTOR_P(0);
|
||||
Vector *b = PG_GETARG_VECTOR_P(1);
|
||||
|
||||
/* TODO Remove in 0.7.0 */
|
||||
CheckDims(a, b);
|
||||
|
||||
PG_RETURN_BOOL(vector_cmp_internal(a, b) > 0);
|
||||
}
|
||||
|
||||
@@ -1225,34 +1160,3 @@ 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;
|
||||
float *values = SPARSEVEC_VALUES(svec);
|
||||
int maxIndex = svec->nnz == 0 ? -1 : svec->indices[svec->nnz - 1];
|
||||
|
||||
if (typmod == -1)
|
||||
dim = maxIndex + 1;
|
||||
else
|
||||
dim = typmod;
|
||||
|
||||
CheckDim(dim);
|
||||
|
||||
if (dim < maxIndex + 1)
|
||||
elog(ERROR, "Vector must have at least %d dimensions", maxIndex + 1);
|
||||
|
||||
result = InitVector(dim);
|
||||
for (int i = 0; i < svec->nnz; i++)
|
||||
result->x[svec->indices[i]] = values[i];
|
||||
|
||||
PG_RETURN_POINTER(result);
|
||||
}
|
||||
|
||||
@@ -1,64 +0,0 @@
|
||||
SELECT hamming_distance(B'111', B'111');
|
||||
hamming_distance
|
||||
------------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance(B'111', B'110');
|
||||
hamming_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance(B'111', B'100');
|
||||
hamming_distance
|
||||
------------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance(B'111', B'000');
|
||||
hamming_distance
|
||||
------------------
|
||||
3
|
||||
(1 row)
|
||||
|
||||
SELECT hamming_distance(B'111', B'00');
|
||||
ERROR: different bit lengths 3 and 2
|
||||
SELECT jaccard_distance(B'1111', B'1111');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'1110');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.25
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'1100');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.5
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'1000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.75
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'0000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1100', B'1000');
|
||||
jaccard_distance
|
||||
------------------
|
||||
0.5
|
||||
(1 row)
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'000');
|
||||
ERROR: different bit lengths 4 and 3
|
||||
@@ -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;
|
||||
|
||||
@@ -24,30 +24,6 @@ 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 < '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector < '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector <= '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector <= '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector = '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
@@ -55,47 +31,7 @@ SELECT '[1,2,3]'::vector = '[1,2,3]';
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector = '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector != '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector != '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector >= '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector >= '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector > '[1,2,3]';
|
||||
?column?
|
||||
----------
|
||||
f
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector > '[1,2]';
|
||||
?column?
|
||||
----------
|
||||
t
|
||||
(1 row)
|
||||
|
||||
ERROR: different vector dimensions 3 and 2
|
||||
SELECT vector_cmp('[1,2,3]', '[1,2,3]');
|
||||
vector_cmp
|
||||
------------
|
||||
@@ -168,152 +104,110 @@ SELECT vector_norm('[3e37,4e37]')::real;
|
||||
5e+37
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[0,0]'::vector, '[3,4]');
|
||||
SELECT l2_distance('[0,0]', '[3,4]');
|
||||
l2_distance
|
||||
-------------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[0,0]'::vector, '[0,1]');
|
||||
SELECT l2_distance('[0,0]', '[0,1]');
|
||||
l2_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('[1,2]'::vector, '[3]');
|
||||
SELECT l2_distance('[1,2]', '[3]');
|
||||
ERROR: different vector dimensions 2 and 1
|
||||
SELECT l2_distance('[3e38]'::vector, '[-3e38]');
|
||||
SELECT l2_distance('[3e38]', '[-3e38]');
|
||||
l2_distance
|
||||
-------------
|
||||
Infinity
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::vector, '[3,4]');
|
||||
SELECT inner_product('[1,2]', '[3,4]');
|
||||
inner_product
|
||||
---------------
|
||||
11
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('[1,2]'::vector, '[3]');
|
||||
SELECT inner_product('[1,2]', '[3]');
|
||||
ERROR: different vector dimensions 2 and 1
|
||||
SELECT inner_product('[3e38]'::vector, '[3e38]');
|
||||
SELECT inner_product('[3e38]', '[3e38]');
|
||||
inner_product
|
||||
---------------
|
||||
Infinity
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::vector, '[2,4]');
|
||||
SELECT cosine_distance('[1,2]', '[2,4]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::vector, '[0,0]');
|
||||
SELECT cosine_distance('[1,2]', '[0,0]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::vector, '[1,1]');
|
||||
SELECT cosine_distance('[1,1]', '[1,1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,0]'::vector, '[0,2]');
|
||||
SELECT cosine_distance('[1,0]', '[0,2]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
|
||||
SELECT cosine_distance('[1,1]', '[-1,-1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,2]'::vector, '[3]');
|
||||
SELECT cosine_distance('[1,2]', '[3]');
|
||||
ERROR: different vector dimensions 2 and 1
|
||||
SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
|
||||
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
|
||||
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('[3e38]'::vector, '[3e38]');
|
||||
SELECT cosine_distance('[3e38]', '[3e38]');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::vector, '[3,4]');
|
||||
SELECT l1_distance('[0,0]', '[3,4]');
|
||||
l1_distance
|
||||
-------------
|
||||
7
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[0,0]'::vector, '[0,1]');
|
||||
SELECT l1_distance('[0,0]', '[0,1]');
|
||||
l1_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT l1_distance('[1,2]'::vector, '[3]');
|
||||
SELECT l1_distance('[1,2]', '[3]');
|
||||
ERROR: different vector dimensions 2 and 1
|
||||
SELECT l1_distance('[3e38]'::vector, '[-3e38]');
|
||||
SELECT l1_distance('[3e38]', '[-3e38]');
|
||||
l1_distance
|
||||
-------------
|
||||
Infinity
|
||||
(1 row)
|
||||
|
||||
SELECT quantize_binary('[1,0,-1]'::vector);
|
||||
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);
|
||||
quantize_binary
|
||||
-----------------
|
||||
01001110101
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);
|
||||
subvector
|
||||
-----------
|
||||
[3,4]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -1, 3);
|
||||
subvector
|
||||
-----------
|
||||
[1]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, 9);
|
||||
subvector
|
||||
-----------
|
||||
[3,4,5]
|
||||
(1 row)
|
||||
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 1, 0);
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, 3, -1);
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT subvector('[1,2,3,4,5]'::vector, -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
|
||||
-----------
|
||||
@@ -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
|
||||
@@ -1,147 +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: value out of range: overflow
|
||||
LINE 1: SELECT '[65520,-65520]'::halfvec;
|
||||
^
|
||||
SELECT '[1e-8,-1e-8]'::halfvec;
|
||||
ERROR: value out of range: underflow
|
||||
LINE 1: SELECT '[1e-8,-1e-8]'::halfvec;
|
||||
^
|
||||
SELECT '[4e38,1]'::halfvec;
|
||||
ERROR: infinite value not allowed in halfvec
|
||||
LINE 1: SELECT '[4e38,1]'::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
|
||||
@@ -1,21 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val bit(3));
|
||||
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
|
||||
INSERT INTO t (val) VALUES (B'110');
|
||||
SELECT * FROM t ORDER BY val <~> B'111';
|
||||
val
|
||||
-----
|
||||
111
|
||||
110
|
||||
100
|
||||
000
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,21 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val bit(4));
|
||||
INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
|
||||
INSERT INTO t (val) VALUES (B'1110');
|
||||
SELECT * FROM t ORDER BY val <%> B'1111';
|
||||
val
|
||||
------
|
||||
1111
|
||||
1110
|
||||
1100
|
||||
0000
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -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;
|
||||
@@ -12,11 +12,14 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
|
||||
[0,0,0]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
|
||||
val
|
||||
---------
|
||||
[0,0,0]
|
||||
[1,1,1]
|
||||
[1,2,3]
|
||||
[1,2,4]
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM t;
|
||||
count
|
||||
@@ -1,26 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
|
||||
val
|
||||
---------------
|
||||
{0:1,1:1,2:1}
|
||||
{0:1,1:2,2:3}
|
||||
{0:1,1:2,2:4}
|
||||
(3 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') 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);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
|
||||
val
|
||||
---------------
|
||||
{0:1,1:2,2:4}
|
||||
{0:1,1:2,2:3}
|
||||
{0:1,1:1,2:1}
|
||||
{}
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,43 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
|
||||
val
|
||||
---------------
|
||||
{0:1,1:2,2:3}
|
||||
{0:1,1:2,2:4}
|
||||
{0:1,1:1,2:1}
|
||||
{}
|
||||
(4 rows)
|
||||
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
|
||||
count
|
||||
-------
|
||||
4
|
||||
(1 row)
|
||||
|
||||
SELECT COUNT(*) FROM t;
|
||||
count
|
||||
-------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
|
||||
val
|
||||
-----
|
||||
(0 rows)
|
||||
|
||||
DROP TABLE t;
|
||||
-- TODO move
|
||||
CREATE TABLE t (val sparsevec);
|
||||
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;
|
||||
@@ -63,25 +63,9 @@ 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
|
||||
LINE 1: SELECT '[-4e38,1]'::vector;
|
||||
^
|
||||
SELECT '[1e-46,1]'::vector;
|
||||
vector
|
||||
--------
|
||||
[0,1]
|
||||
(1 row)
|
||||
|
||||
SELECT '[-1e-46,1]'::vector;
|
||||
vector
|
||||
--------
|
||||
[-0,1]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3'::vector;
|
||||
ERROR: malformed vector literal: "[1,2,3"
|
||||
LINE 1: SELECT '[1,2,3'::vector;
|
||||
@@ -132,30 +116,8 @@ SELECT '[1, ,3]'::vector;
|
||||
ERROR: invalid input syntax for type vector: "[1, ,3]"
|
||||
LINE 1: SELECT '[1, ,3]'::vector;
|
||||
^
|
||||
SELECT '[1,2,3]'::vector(3);
|
||||
vector
|
||||
---------
|
||||
[1,2,3]
|
||||
(1 row)
|
||||
|
||||
SELECT '[1,2,3]'::vector(2);
|
||||
ERROR: expected 2 dimensions, not 3
|
||||
SELECT '[1,2,3]'::vector(3, 2);
|
||||
ERROR: invalid type modifier
|
||||
LINE 1: SELECT '[1,2,3]'::vector(3, 2);
|
||||
^
|
||||
SELECT '[1,2,3]'::vector('a');
|
||||
ERROR: invalid input syntax for type integer: "a"
|
||||
LINE 1: SELECT '[1,2,3]'::vector('a');
|
||||
^
|
||||
SELECT '[1,2,3]'::vector(0);
|
||||
ERROR: dimensions for type vector must be at least 1
|
||||
LINE 1: SELECT '[1,2,3]'::vector(0);
|
||||
^
|
||||
SELECT '[1,2,3]'::vector(16001);
|
||||
ERROR: dimensions for type vector cannot exceed 16000
|
||||
LINE 1: SELECT '[1,2,3]'::vector(16001);
|
||||
^
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
|
||||
unnest
|
||||
---------
|
||||
@@ -1,60 +0,0 @@
|
||||
SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
|
||||
l2_distance
|
||||
-------------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT l2_distance('{}'::sparsevec, '{1:1}');
|
||||
l2_distance
|
||||
-------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT '{}'::sparsevec <-> '{0:3,1:4}';
|
||||
?column?
|
||||
----------
|
||||
5
|
||||
(1 row)
|
||||
|
||||
SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
|
||||
inner_product
|
||||
---------------
|
||||
10
|
||||
(1 row)
|
||||
|
||||
SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
|
||||
sparsevec_negative_inner_product
|
||||
----------------------------------
|
||||
-10
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
|
||||
cosine_distance
|
||||
-----------------
|
||||
0
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
|
||||
cosine_distance
|
||||
-----------------
|
||||
2
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
|
||||
cosine_distance
|
||||
-----------------
|
||||
1
|
||||
(1 row)
|
||||
|
||||
SELECT cosine_distance('{}'::sparsevec, '{}');
|
||||
cosine_distance
|
||||
-----------------
|
||||
NaN
|
||||
(1 row)
|
||||
|
||||
@@ -1,64 +0,0 @@
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec;
|
||||
sparsevec
|
||||
---------------
|
||||
{0:1.5,2:3.5}
|
||||
(1 row)
|
||||
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
|
||||
vector
|
||||
-------------
|
||||
[1.5,0,3.5]
|
||||
(1 row)
|
||||
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
|
||||
vector
|
||||
-----------------
|
||||
[1.5,0,3.5,0,0]
|
||||
(1 row)
|
||||
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
|
||||
vector
|
||||
---------------
|
||||
[1.5,0,3.5,0]
|
||||
(1 row)
|
||||
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
|
||||
ERROR: Vector must have at least 3 dimensions
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
sparsevec
|
||||
---------------
|
||||
{1:1.5,3:3.5}
|
||||
(1 row)
|
||||
|
||||
SELECT '{0:0,1:1,2:0}'::sparsevec;
|
||||
sparsevec
|
||||
-----------
|
||||
{1:1}
|
||||
(1 row)
|
||||
|
||||
SELECT '{1:1,0:1}'::sparsevec;
|
||||
ERROR: indexes must be in ascending order
|
||||
LINE 1: SELECT '{1:1,0:1}'::sparsevec;
|
||||
^
|
||||
SELECT '{}'::sparsevec;
|
||||
sparsevec
|
||||
-----------
|
||||
{}
|
||||
(1 row)
|
||||
|
||||
SELECT '{}'::sparsevec::vector;
|
||||
ERROR: vector must have at least 1 dimension
|
||||
SELECT '{-1:1}'::sparsevec;
|
||||
ERROR: index "-1" is out of range for type sparsevec
|
||||
LINE 1: SELECT '{-1:1}'::sparsevec;
|
||||
^
|
||||
SELECT '{1:1}'::sparsevec;
|
||||
sparsevec
|
||||
-----------
|
||||
{1:1}
|
||||
(1 row)
|
||||
|
||||
SELECT '{}'::sparsevec(2);
|
||||
ERROR: type modifier is not allowed for type "sparsevec"
|
||||
LINE 1: SELECT '{}'::sparsevec(2);
|
||||
^
|
||||
@@ -1,13 +0,0 @@
|
||||
SELECT hamming_distance(B'111', B'111');
|
||||
SELECT hamming_distance(B'111', B'110');
|
||||
SELECT hamming_distance(B'111', B'100');
|
||||
SELECT hamming_distance(B'111', B'000');
|
||||
SELECT hamming_distance(B'111', B'00');
|
||||
|
||||
SELECT jaccard_distance(B'1111', B'1111');
|
||||
SELECT jaccard_distance(B'1111', B'1110');
|
||||
SELECT jaccard_distance(B'1111', B'1100');
|
||||
SELECT jaccard_distance(B'1111', B'1000');
|
||||
SELECT jaccard_distance(B'1111', B'0000');
|
||||
SELECT jaccard_distance(B'1100', B'1000');
|
||||
SELECT jaccard_distance(B'1111', B'000');
|
||||
@@ -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)
|
||||
|
||||
62
test/sql/functions.sql
Normal file
62
test/sql/functions.sql
Normal file
@@ -0,0 +1,62 @@
|
||||
SELECT '[1,2,3]'::vector + '[4,5,6]';
|
||||
SELECT '[3e38]'::vector + '[3e38]';
|
||||
SELECT '[1,2,3]'::vector - '[4,5,6]';
|
||||
SELECT '[-3e38]'::vector - '[3e38]';
|
||||
SELECT '[1,2,3]'::vector * '[4,5,6]';
|
||||
SELECT '[1e37]'::vector * '[1e37]';
|
||||
SELECT '[1e-37]'::vector * '[1e-37]';
|
||||
|
||||
SELECT '[1,2,3]'::vector = '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector = '[1,2]';
|
||||
|
||||
SELECT vector_cmp('[1,2,3]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2,3]', '[0,0,0]');
|
||||
SELECT vector_cmp('[0,0,0]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2,3]', '[1,2]');
|
||||
SELECT vector_cmp('[1,2]', '[2,3,4]');
|
||||
SELECT vector_cmp('[2,3]', '[1,2,3]');
|
||||
|
||||
SELECT vector_dims('[1,2,3]');
|
||||
|
||||
SELECT round(vector_norm('[1,1]')::numeric, 5);
|
||||
SELECT vector_norm('[3,4]');
|
||||
SELECT vector_norm('[0,1]');
|
||||
SELECT vector_norm('[3e37,4e37]')::real;
|
||||
|
||||
SELECT l2_distance('[0,0]', '[3,4]');
|
||||
SELECT l2_distance('[0,0]', '[0,1]');
|
||||
SELECT l2_distance('[1,2]', '[3]');
|
||||
SELECT l2_distance('[3e38]', '[-3e38]');
|
||||
|
||||
SELECT inner_product('[1,2]', '[3,4]');
|
||||
SELECT inner_product('[1,2]', '[3]');
|
||||
SELECT inner_product('[3e38]', '[3e38]');
|
||||
|
||||
SELECT cosine_distance('[1,2]', '[2,4]');
|
||||
SELECT cosine_distance('[1,2]', '[0,0]');
|
||||
SELECT cosine_distance('[1,1]', '[1,1]');
|
||||
SELECT cosine_distance('[1,0]', '[0,2]');
|
||||
SELECT cosine_distance('[1,1]', '[-1,-1]');
|
||||
SELECT cosine_distance('[1,2]', '[3]');
|
||||
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
|
||||
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
|
||||
SELECT cosine_distance('[3e38]', '[3e38]');
|
||||
|
||||
SELECT l1_distance('[0,0]', '[3,4]');
|
||||
SELECT l1_distance('[0,0]', '[0,1]');
|
||||
SELECT l1_distance('[1,2]', '[3]');
|
||||
SELECT l1_distance('[3e38]', '[-3e38]');
|
||||
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
|
||||
SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
|
||||
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
|
||||
@@ -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);
|
||||
@@ -1,34 +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 '[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)[];
|
||||
@@ -1,12 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val bit(3));
|
||||
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES (B'110');
|
||||
|
||||
SELECT * FROM t ORDER BY val <~> B'111';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,12 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val bit(4));
|
||||
INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES (B'1110');
|
||||
|
||||
SELECT * FROM t ORDER BY val <%> B'1111';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -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;
|
||||
@@ -7,7 +7,7 @@ CREATE INDEX ON t USING hnsw (val vector_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::vector)) t2;
|
||||
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
|
||||
SELECT COUNT(*) FROM t;
|
||||
|
||||
TRUNCATE t;
|
||||
@@ -1,13 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
|
||||
SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') t2;
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,12 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
|
||||
SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
|
||||
|
||||
DROP TABLE t;
|
||||
@@ -1,25 +0,0 @@
|
||||
SET enable_seqscan = off;
|
||||
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
|
||||
INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
|
||||
|
||||
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
|
||||
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
|
||||
SELECT COUNT(*) FROM t;
|
||||
|
||||
TRUNCATE t;
|
||||
SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
|
||||
|
||||
DROP TABLE t;
|
||||
|
||||
-- TODO move
|
||||
CREATE TABLE t (val sparsevec);
|
||||
INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
|
||||
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
|
||||
TRUNCATE t;
|
||||
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;
|
||||
@@ -11,9 +11,6 @@ SELECT '[1.5e38,-1.5e38]'::vector;
|
||||
SELECT '[1.5e+38,-1.5e+38]'::vector;
|
||||
SELECT '[1.5e-38,-1.5e-38]'::vector;
|
||||
SELECT '[4e38,1]'::vector;
|
||||
SELECT '[-4e38,1]'::vector;
|
||||
SELECT '[1e-46,1]'::vector;
|
||||
SELECT '[-1e-46,1]'::vector;
|
||||
SELECT '[1,2,3'::vector;
|
||||
SELECT '[1,2,3]9'::vector;
|
||||
SELECT '1,2,3'::vector;
|
||||
@@ -25,13 +22,7 @@ SELECT '[1,]'::vector;
|
||||
SELECT '[1a]'::vector;
|
||||
SELECT '[1,,3]'::vector;
|
||||
SELECT '[1, ,3]'::vector;
|
||||
|
||||
SELECT '[1,2,3]'::vector(3);
|
||||
SELECT '[1,2,3]'::vector(2);
|
||||
SELECT '[1,2,3]'::vector(3, 2);
|
||||
SELECT '[1,2,3]'::vector('a');
|
||||
SELECT '[1,2,3]'::vector(0);
|
||||
SELECT '[1,2,3]'::vector(16001);
|
||||
|
||||
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
|
||||
SELECT '{"[1,2,3]"}'::vector(2)[];
|
||||
@@ -1,12 +0,0 @@
|
||||
SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
|
||||
SELECT l2_distance('{}'::sparsevec, '{1:1}');
|
||||
SELECT '{}'::sparsevec <-> '{0:3,1:4}';
|
||||
|
||||
SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
|
||||
SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
|
||||
|
||||
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
|
||||
SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
|
||||
SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
|
||||
SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
|
||||
SELECT cosine_distance('{}'::sparsevec, '{}');
|
||||
@@ -1,18 +0,0 @@
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec;
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
|
||||
SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
|
||||
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
|
||||
|
||||
SELECT '{0:0,1:1,2:0}'::sparsevec;
|
||||
|
||||
SELECT '{1:1,0:1}'::sparsevec;
|
||||
|
||||
SELECT '{}'::sparsevec;
|
||||
SELECT '{}'::sparsevec::vector;
|
||||
|
||||
SELECT '{-1:1}'::sparsevec;
|
||||
SELECT '{1:1}'::sparsevec;
|
||||
|
||||
SELECT '{}'::sparsevec(2);
|
||||
@@ -1,83 +0,0 @@
|
||||
SELECT '[1,2,3]'::vector + '[4,5,6]';
|
||||
SELECT '[3e38]'::vector + '[3e38]';
|
||||
SELECT '[1,2,3]'::vector - '[4,5,6]';
|
||||
SELECT '[-3e38]'::vector - '[3e38]';
|
||||
SELECT '[1,2,3]'::vector * '[4,5,6]';
|
||||
SELECT '[1e37]'::vector * '[1e37]';
|
||||
SELECT '[1e-37]'::vector * '[1e-37]';
|
||||
|
||||
SELECT '[1,2,3]'::vector < '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector < '[1,2]';
|
||||
SELECT '[1,2,3]'::vector <= '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector <= '[1,2]';
|
||||
SELECT '[1,2,3]'::vector = '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector = '[1,2]';
|
||||
SELECT '[1,2,3]'::vector != '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector != '[1,2]';
|
||||
SELECT '[1,2,3]'::vector >= '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector >= '[1,2]';
|
||||
SELECT '[1,2,3]'::vector > '[1,2,3]';
|
||||
SELECT '[1,2,3]'::vector > '[1,2]';
|
||||
|
||||
SELECT vector_cmp('[1,2,3]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2,3]', '[0,0,0]');
|
||||
SELECT vector_cmp('[0,0,0]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2]', '[1,2,3]');
|
||||
SELECT vector_cmp('[1,2,3]', '[1,2]');
|
||||
SELECT vector_cmp('[1,2]', '[2,3,4]');
|
||||
SELECT vector_cmp('[2,3]', '[1,2,3]');
|
||||
|
||||
SELECT vector_dims('[1,2,3]');
|
||||
|
||||
SELECT round(vector_norm('[1,1]')::numeric, 5);
|
||||
SELECT vector_norm('[3,4]');
|
||||
SELECT vector_norm('[0,1]');
|
||||
SELECT vector_norm('[3e37,4e37]')::real;
|
||||
|
||||
SELECT l2_distance('[0,0]'::vector, '[3,4]');
|
||||
SELECT l2_distance('[0,0]'::vector, '[0,1]');
|
||||
SELECT l2_distance('[1,2]'::vector, '[3]');
|
||||
SELECT l2_distance('[3e38]'::vector, '[-3e38]');
|
||||
|
||||
SELECT inner_product('[1,2]'::vector, '[3,4]');
|
||||
SELECT inner_product('[1,2]'::vector, '[3]');
|
||||
SELECT inner_product('[3e38]'::vector, '[3e38]');
|
||||
|
||||
SELECT cosine_distance('[1,2]'::vector, '[2,4]');
|
||||
SELECT cosine_distance('[1,2]'::vector, '[0,0]');
|
||||
SELECT cosine_distance('[1,1]'::vector, '[1,1]');
|
||||
SELECT cosine_distance('[1,0]'::vector, '[0,2]');
|
||||
SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
|
||||
SELECT cosine_distance('[1,2]'::vector, '[3]');
|
||||
SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
|
||||
SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
|
||||
SELECT cosine_distance('[3e38]'::vector, '[3e38]');
|
||||
|
||||
SELECT l1_distance('[0,0]'::vector, '[3,4]');
|
||||
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 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 avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
|
||||
SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
|
||||
SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
|
||||
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
|
||||
SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
|
||||
@@ -86,7 +86,7 @@ foreach (@queries)
|
||||
push(@expected, $res);
|
||||
}
|
||||
|
||||
test_recall(0.19, $limit, "before vacuum");
|
||||
test_recall(0.20, $limit, "before vacuum");
|
||||
test_recall(0.95, 100, "before vacuum");
|
||||
|
||||
# TODO Test concurrent inserts with vacuum
|
||||
|
||||
113
test/t/020_hnsw_array.pl
Normal file
113
test/t/020_hnsw_array.pl
Normal file
@@ -0,0 +1,113 @@
|
||||
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 float4[3]);");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
$node->safe_psql("postgres", qq(
|
||||
CREATE FUNCTION float4_l2_distance(float4[], float4[]) RETURNS float8
|
||||
AS 'BEGIN RETURN l2_distance(\$1::vector, \$2::vector); END;'
|
||||
LANGUAGE plpgsql IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE FUNCTION float4_l2_squared_distance(float4[], float4[]) RETURNS float8
|
||||
AS 'BEGIN RETURN vector_l2_squared_distance(\$1::vector, \$2::vector); END;'
|
||||
LANGUAGE plpgsql IMMUTABLE STRICT PARALLEL SAFE;
|
||||
|
||||
CREATE OPERATOR <-> (
|
||||
LEFTARG = float4[], RIGHTARG = float4[], PROCEDURE = float4_l2_distance,
|
||||
COMMUTATOR = '<->'
|
||||
);
|
||||
|
||||
CREATE OPERATOR CLASS float4_l2_ops
|
||||
FOR TYPE float4[] USING hnsw AS
|
||||
OPERATOR 1 <-> (float4[], float4[]) FOR ORDER BY float_ops,
|
||||
FUNCTION 1 float4_l2_squared_distance(float4[], float4[]);
|
||||
));
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my $r1 = rand();
|
||||
my $r2 = rand();
|
||||
my $r3 = rand();
|
||||
push(@queries, "{$r1,$r2,$r3}");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<->");
|
||||
my @opclasses = ("float4_l2_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);
|
||||
}
|
||||
|
||||
# Add index
|
||||
$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v $opclass);");
|
||||
|
||||
my $min = $operator eq "<#>" ? 0.80 : 0.99;
|
||||
test_recall($min, $operator);
|
||||
}
|
||||
|
||||
done_testing();
|
||||
@@ -1,137 +0,0 @@
|
||||
use strict;
|
||||
use warnings;
|
||||
use PostgresNode;
|
||||
use TestLib;
|
||||
use Test::More;
|
||||
|
||||
my $node;
|
||||
my @queries = ();
|
||||
my @expected;
|
||||
my $limit = 20;
|
||||
my $dim = 52;
|
||||
my $max = 2**$dim;
|
||||
|
||||
sub test_recall
|
||||
{
|
||||
my ($min, $operator) = @_;
|
||||
my $correct = 0;
|
||||
my $total = 0;
|
||||
|
||||
my $explain = $node->safe_psql("postgres", qq(
|
||||
SET enable_seqscan = off;
|
||||
SET hnsw.ef_search = 100;
|
||||
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;
|
||||
SET hnsw.ef_search = 100;
|
||||
SELECT i FROM tst ORDER BY v $operator $queries[$i] LIMIT $limit;
|
||||
));
|
||||
my @actual_ids = split("\n", $actual);
|
||||
|
||||
my @expected_ids = split("\n", $expected[$i]);
|
||||
my %expected_set = map { $_ => 1 } @expected_ids;
|
||||
|
||||
foreach (@actual_ids)
|
||||
{
|
||||
if (exists($expected_set{$_}))
|
||||
{
|
||||
$correct++;
|
||||
}
|
||||
}
|
||||
|
||||
$total += $limit;
|
||||
}
|
||||
|
||||
cmp_ok($correct / $total, ">=", $min, $operator);
|
||||
}
|
||||
|
||||
# Initialize node
|
||||
$node = get_new_node('node');
|
||||
$node->init;
|
||||
$node->start;
|
||||
|
||||
# Create table
|
||||
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
|
||||
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v bit($dim));");
|
||||
$node->safe_psql("postgres",
|
||||
"INSERT INTO tst SELECT i, (random() * $max)::bigint::bit($dim) FROM generate_series(1, 10000) i;"
|
||||
);
|
||||
|
||||
# Generate queries
|
||||
for (1 .. 20)
|
||||
{
|
||||
my $r = int(rand() * $max);
|
||||
push(@queries, "${r}::bigint::bit($dim)");
|
||||
}
|
||||
|
||||
# Check each index type
|
||||
my @operators = ("<~>", "<\%>");
|
||||
my @opclasses = ("bit_hamming_ops", "bit_jaccard_ops");
|
||||
|
||||
for my $i (0 .. $#operators)
|
||||
{
|
||||
my $operator = $operators[$i];
|
||||
my $opclass = $opclasses[$i];
|
||||
|
||||
# Get exact results
|
||||
@expected = ();
|
||||
foreach (@queries)
|
||||
{
|
||||
# Handle ties
|
||||
my $res = $node->safe_psql("postgres", qq(
|
||||
WITH top AS (
|
||||
SELECT v $operator $_ AS distance FROM tst ORDER BY v $operator $_ LIMIT $limit
|
||||
)
|
||||
SELECT i FROM tst WHERE (v $operator $_) <= (SELECT MAX(distance) FROM top)
|
||||
));
|
||||
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 = $operator eq "<\%>" ? 0.96 : 0.98;
|
||||
test_recall($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();
|
||||
@@ -1,132 +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/);
|
||||
|
||||
$node->safe_psql("postgres", "DROP INDEX idx;");
|
||||
}
|
||||
|
||||
done_testing();
|
||||
@@ -1,4 +1,4 @@
|
||||
comment = 'vector data type and ivfflat and hnsw access methods'
|
||||
default_version = '0.6.2'
|
||||
default_version = '0.6.0'
|
||||
module_pathname = '$libdir/vector'
|
||||
relocatable = true
|
||||
|
||||
Reference in New Issue
Block a user