NGT - high-speed approximate nearest neighbors - for PHP
Run:
composer require ankane/ngtAdd scripts to composer.json to download the shared library:
"scripts": {
"post-install-cmd": "Ngt\\Vendor::check",
"post-update-cmd": "Ngt\\Vendor::check"
}And run:
composer installOn Mac, also install OpenMP:
brew install libompNGT is not available for Windows
Prep your data
$objects = [
[1, 1, 2, 1],
[5, 4, 6, 5],
[1, 2, 1, 2]
];Create an index
$index = new Ngt\Index($dimensions);Insert objects
$index->batchInsert($objects);Search the index
$index->search($query, size: 3);Save the index
$index->save($path);Load an index
$index = Ngt\Index::load($path);Get an object by id
$index->object($id);Insert a single object
$index->insert($object);Remove an object by id
$index->remove($id);Build the index
$index->buildIndex();$dim = 10;
$objects = [];
for ($i = 0; $i < 100; $i++) {
$object = [];
for ($j = 0; $j < $dim; $j++) {
$object[] = rand(0, 100);
}
$objects[] = $object;
}
$index = new Ngt\Index($dim);
$index->batchInsert($objects);
$query = $objects[0];
$result = $index->search($query, size: 3);
foreach ($result as $res) {
print($res['id'] . ', ' . $res['distance'] . "\n");
}Defaults shown below
use Ngt\DistanceType;
use Ngt\ObjectType;
new Ngt\Index(
$dimensions,
edgeSizeForCreation: 10,
edgeSizeForSearch: 40,
distanceType: DistanceType::L2, // L1, L2, Hamming, Angle, Cosine, NormalizedAngle, NormalizedCosine, Jaccard
objectType: ObjectType::Float // Float, Float16, Integer
);This library is modeled after NGT’s Python API.
View the changelog
Everyone is encouraged to help improve this project. Here are a few ways you can help:
- Report bugs
- Fix bugs and submit pull requests
- Write, clarify, or fix documentation
- Suggest or add new features
To get started with development:
git clone https://github.com/ankane/ngt-php.git
cd ngt-php
composer install
composer test