usearch 2.15.3 → 2.16.0

package/javascript/usearch.test.js

@@ -1,5 +1,8 @@
 const test = require('node:test');
 const assert = require('node:assert');
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
 const usearch = require('./dist/cjs/usearch.js');
 
 function assertAlmostEqual(actual, expected, tolerance = 1e-6) {
@@ -12,38 +15,84 @@ function assertAlmostEqual(actual, expected, tolerance = 1e-6) {
 }
 
 
-test('Single-entry operations', () => {
-    const index = new usearch.Index(2, 'l2sq');
+test('Single-entry operations', async (t) => {
+    await t.test('index info', () => {
+        const index = new usearch.Index(2, 'l2sq');
 
-    assert.equal(index.connectivity(), 16, 'connectivity should be 16');
-    assert.equal(index.dimensions(), 2, 'dimensions should be 2');
-    assert.equal(index.size(), 0, 'initial size should be 0');
+        assert.equal(index.connectivity(), 16, 'connectivity should be 16');
+        assert.equal(index.dimensions(), 2, 'dimensions should be 2');
+        assert.equal(index.size(), 0, 'initial size should be 0');
+    });
+
+    await t.test('add', () => {
+        const index = new usearch.Index(2, 'l2sq');
+
+        index.add(15n, new Float32Array([10, 20]));
+        index.add(16n, new Float32Array([10, 25]));
+
+        assert.equal(index.size(), 2, 'size after adding elements should be 2');
+        assert.equal(index.contains(15), true, 'entry must be present after insertion');
+
+        const results = index.search(new Float32Array([13, 14]), 2);
+
+        assert.deepEqual(results.keys, new BigUint64Array([15n, 16n]), 'keys should be 15 and 16');
+        assert.deepEqual(results.distances, new Float32Array([45, 130]), 'distances should be 45 and 130');
+    });
+
+    await t.test('remove', () => {
+        const index = new usearch.Index(2, 'l2sq');
+
+        index.add(15n, new Float32Array([10, 20]));
+        index.add(16n, new Float32Array([10, 25]));
+        index.add(25n, new Float32Array([20, 40]));
+        index.add(26n, new Float32Array([20, 45]));
 
-    index.add(15n, new Float32Array([10, 20]));
-    index.add(16n, new Float32Array([10, 25]));
+        assert.equal(index.remove(15n), 1);
 
-    assert.equal(index.size(), 2, 'size after adding elements should be 2');
-    assert.equal(index.contains(15), true, 'entry must be present after insertion');
+        assert.equal(index.size(), 3, 'size after remoing elements should be 3');
+        assert.equal(index.contains(15), false, 'entry must be absent after insertion');
 
-    const results = index.search(new Float32Array([13, 14]), 2);
+        const results = index.search(new Float32Array([13, 14]), 2);
 
-    assert.deepEqual(results.keys, new BigUint64Array([15n, 16n]), 'keys should be 15 and 16');
-    assert.deepEqual(results.distances, new Float32Array([45, 130]), 'distances should be 45 and 130');
+        assert.deepEqual(results.keys, new BigUint64Array([16n, 25n]), 'keys should not include 15');
+    });
 });
 
-test('Batch operations', () => {
-    const indexBatch = new usearch.Index(2, 'l2sq');
+test('Batch operations', async (t) => {
+    await t.test('add', () => {
+        const indexBatch = new usearch.Index(2, 'l2sq');
 
-    const keys = [15n, 16n];
-    const vectors = [new Float32Array([10, 20]), new Float32Array([10, 25])];
+        const keys = [15n, 16n];
+        const vectors = [new Float32Array([10, 20]), new Float32Array([10, 25])];
 
-    indexBatch.add(keys, vectors);
-    assert.equal(indexBatch.size(), 2, 'size after adding batch should be 2');
+        indexBatch.add(keys, vectors);
+        assert.equal(indexBatch.size(), 2, 'size after adding batch should be 2');
 
-    const results = indexBatch.search(new Float32Array([13, 14]), 2);
+        const results = indexBatch.search(new Float32Array([13, 14]), 2);
 
-    assert.deepEqual(results.keys, new BigUint64Array([15n, 16n]), 'keys should be 15 and 16');
-    assert.deepEqual(results.distances, new Float32Array([45, 130]), 'distances should be 45 and 130');
+        assert.deepEqual(results.keys, new BigUint64Array([15n, 16n]), 'keys should be 15 and 16');
+        assert.deepEqual(results.distances, new Float32Array([45, 130]), 'distances should be 45 and 130');
+    });
+
+    await t.test('remove', () => {
+        const indexBatch = new usearch.Index(2, 'l2sq');
+
+        const keys = [15n, 16n, 25n, 26n];
+        const vectors = [
+            new Float32Array([10, 20]),
+            new Float32Array([10, 25]),
+            new Float32Array([20, 40]),
+            new Float32Array([20, 45])
+        ];
+        indexBatch.add(keys, vectors);
+
+        assert.deepEqual(indexBatch.remove([15n, 25n]), [1, 1])
+        assert.equal(indexBatch.size(), 2, 'size after removing batch should be 2');
+
+        const results = indexBatch.search(new Float32Array([13, 14]), 2);
+
+        assert.deepEqual(results.keys, new BigUint64Array([16n, 26n]), 'keys should not include 15 and 25');
+    });
 });
 
 test("Expected results", () => {
@@ -60,7 +109,27 @@ test("Expected results", () => {
     assertAlmostEqual(results.distances[0], new Float32Array([0]));
 });
 
+test('Expected count()', async (t) => {
+    const index = new usearch.Index({
+        metric: 'l2sq',
+        connectivity: 16,
+        dimensions: 3,
+    });
+    index.add(
+        [42n, 43n],
+        [new Float32Array([0.2, 0.6, 0.4]), new Float32Array([0.2, 0.6, 0.4])]
+    );
 
+    await t.test('Argument is a number', () => {
+        assert.equal(1, index.count(43n));
+    });
+    await t.test('Argument is a number (does not exist)', () => {
+        assert.equal(0, index.count(44n));
+    });
+    await t.test('Argument is an array', () => {
+        assert.deepEqual([1, 1, 0], index.count([42n, 43n, 44n]));
+    });
+});
 
 test('Operations with invalid values', () => {
     const indexBatch = new usearch.Index(2, 'l2sq');
@@ -84,3 +153,103 @@ test('Operations with invalid values', () => {
         }
     );
 });
+
+test('Invalid operations', async (t) => {
+    await t.test('Add the same keys', () => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.add(42n, new Float32Array([0.2, 0.6, 0.4]));
+        assert.throws(
+            () => index.add(42n, new Float32Array([0.2, 0.6, 0.4])),
+            {
+                name: 'Error',
+                message: 'Duplicate keys not allowed in high-level wrappers'
+            }
+        );
+    });
+});
+
+
+test('Serialization', async (t) => {
+    const indexPath = path.join(os.tmpdir(), 'usearch.test.index')
+
+    t.beforeEach(() => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.add(42n, new Float32Array([0.2, 0.6, 0.4]));
+        index.save(indexPath);
+    });
+
+    t.afterEach(() => {
+        fs.unlinkSync(indexPath);
+    });
+
+    await t.test('load', () => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.load(indexPath);
+        const results = index.search(new Float32Array([0.2, 0.6, 0.4]), 10);
+
+        assert.equal(index.size(), 1);
+        assert.deepEqual(results.keys, new BigUint64Array([42n]));
+        assertAlmostEqual(results.distances[0], new Float32Array([0]));
+    });
+
+    // todo: Skip as the test fails only on windows.
+    // The following error in afterEach().
+    // `error: "EBUSY: resource busy or locked, unlink`
+    await t.test('view: Read data', {skip: process.platform === 'win32'}, () => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.view(indexPath);
+        const results = index.search(new Float32Array([0.2, 0.6, 0.4]), 10);
+
+        assert.equal(index.size(), 1);
+        assert.deepEqual(results.keys, new BigUint64Array([42n]));
+        assertAlmostEqual(results.distances[0], new Float32Array([0]));
+    });
+
+    await t.test('view: Invalid operations: add', {skip: process.platform === 'win32'}, () => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.view(indexPath);
+        assert.throws(
+            () => index.add(43n, new Float32Array([0.2, 0.6, 0.4])),
+            {
+                name: 'Error',
+                message: "Can't add to an immutable index"
+            }
+        );
+    });
+
+    await t.test('view: Invalid operations: remove', {skip: process.platform === 'win32'}, () => {
+        const index = new usearch.Index({
+            metric: "l2sq",
+            connectivity: 16,
+            dimensions: 3,
+        });
+        index.view(indexPath);
+        assert.throws(
+            () => index.remove(42n),
+            {
+                name: 'Error',
+                message: "Can't remove from an immutable index"
+            }
+        );
+    });
+});

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "usearch",
-    "version": "2.15.3",
+    "version": "2.16.0",
     "description": "Smaller & Faster Single-File Vector Search Engine from Unum",
     "author": "Ash Vardanian (https://ashvardanian.com/)",
     "license": "Apache 2.0",

package/simsimd/.git-blame-ignore-revs

@@ -0,0 +1,2 @@
+a4022a988287e527757ecc9bc16a4f2e7dc4770e
+750c59f5116a2000507a0cec09db009fd7d31232

package/simsimd/CONTRIBUTING.md

@@ -1,5 +1,11 @@
 # Contributing
 
+To keep the quality of the code high, we have a set of [guidelines](https://github.com/unum-cloud).
+
+- [What's the procedure?](https://github.com/unum-cloud/awesome/blob/main/Workflow.md#organizing-software-development)
+- [How to organize branches?](https://github.com/unum-cloud/awesome/blob/main/Workflow.md#branches)
+- [How to style commits?](https://github.com/unum-cloud/awesome/blob/main/Workflow.md#commits)
+
 ## C and C++
 
 To rerun experiments utilize the following command:
@@ -23,15 +29,37 @@ sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 100
 sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-12 100
 ```
 
-On MacOS it's recommended to use Homebrew and install Clang, as opposed to "Apple Clang".
-Replacing the default compiler is not recommended, as it may break the system, but you can pass it as an environment variable:
+To compile with the default Apple Clang on MacOS, use:
 
 ```sh
-brew install llvm
-cmake -D CMAKE_BUILD_TYPE=Release -D SIMSIMD_BUILD_TESTS=1 \
-    -D CMAKE_C_COMPILER="$(brew --prefix llvm)/bin/clang" \
-    -D CMAKE_CXX_COMPILER="$(brew --prefix llvm)/bin/clang++" \
-    -B build_release
+brew install openblas
+cmake -D CMAKE_BUILD_TYPE=Release \
+      -D SIMSIMD_BUILD_TESTS=1 \
+      -D SIMSIMD_BUILD_BENCHMARKS=1 \
+      -D SIMSIMD_BUILD_BENCHMARKS_WITH_CBLAS=1 \
+      -D CMAKE_PREFIX_PATH="$(brew --prefix openblas)" \
+      -D CMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES="$(brew --prefix openblas)/include" \
+      -B build_release
+cmake --build build_release --config Release
+```
+
+On MacOS it's recommended to use Homebrew and install Clang, as opposed to "Apple Clang".
+Replacing the default compiler across the entire system is not recommended on MacOS, as it may break the system, but you can pass it as an environment variable:
+
+```sh
+brew install llvm openblas
+cmake -D CMAKE_BUILD_TYPE=Release \
+      -D SIMSIMD_BUILD_TESTS=1 \
+      -D SIMSIMD_BUILD_BENCHMARKS=1 \
+      -D SIMSIMD_BUILD_BENCHMARKS_WITH_CBLAS=1 \
+      -D CMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES="$(brew --prefix openblas)/include" \
+      -D CMAKE_C_LINK_FLAGS="-L$(xcrun --sdk macosx --show-sdk-path)/usr/lib" \
+      -D CMAKE_EXE_LINKER_FLAGS="-L$(xcrun --sdk macosx --show-sdk-path)/usr/lib" \
+      -D CMAKE_C_COMPILER="$(brew --prefix llvm)/bin/clang" \
+      -D CMAKE_CXX_COMPILER="$(brew --prefix llvm)/bin/clang++" \
+      -D CMAKE_OSX_SYSROOT="$(xcrun --sdk macosx --show-sdk-path)" \
+      -D CMAKE_OSX_DEPLOYMENT_TARGET=$(sw_vers -productVersion) \
+      -B build_release
 cmake --build build_release --config Release
 ```
 
@@ -42,7 +70,7 @@ Testing:
 ```sh
 pip install -e .                             # to install the package in editable mode
 pip install pytest pytest-repeat tabulate    # testing dependencies
-pytest python/test.py -s -x -Wd              # to run tests
+pytest scripts/test.py -s -x -Wd             # to run tests
 
 # to check supported SIMD instructions:
 python -c "import simsimd; print(simsimd.get_capabilities())" 
@@ -62,28 +90,42 @@ CC=$(brew --prefix llvm)/bin/clang CXX=$(brew --prefix llvm)/bin/clang++ pip ins
 Benchmarking:
 
 ```sh
-pip install numpy scipy scikit-learn        # for comparison baselines
-python python/bench.py                      # to run default benchmarks
-python python/bench.py --n 1000 --ndim 1536 # batch size and dimensions
+pip install numpy scipy scikit-learn                 # for comparison baselines
+python scripts/bench_vectors.py                      # to run default benchmarks
+python scripts/bench_vectors.py --n 1000 --ndim 1536 # batch size and dimensions
 ```
 
-You can also benchmark against other libraries:
+You can also benchmark against other libraries, filter the numeric types, and distance metrics:
 
 ```sh
-$ python python/bench.py --help
-> usage: bench.py [-h] [--n N] [--ndim NDIM] [--scipy] [--scikit] [--torch] [--tf] [--jax]
+$ python scripts/bench_vectors.py --help
+> usage: bench.py [-h] [--ndim NDIM] [-n COUNT]
+>                 [--metric {all,dot,spatial,binary,probability,sparse}]
+>                 [--dtype {all,bits,int8,uint16,uint32,float16,float32,float64,bfloat16,complex32,complex64,complex128}] 
+>                 [--scipy] [--scikit] [--torch] [--tf] [--jax]
 > 
 > Benchmark SimSIMD vs. other libraries
 > 
-> options:
->   -h, --help   show this help message and exit
->   --n N        Number of vectors (default: 1000)
->   --ndim NDIM  Number of dimensions (default: 1536)
->   --scipy      Profile SciPy
->   --scikit     Profile scikit-learn
->   --torch      Profile PyTorch
->   --tf         Profile TensorFlow
->   --jax        Profile JAX
+> optional arguments:
+>   -h, --help            show this help message and exit
+>   --ndim NDIM           Number of dimensions in vectors (default: 1536) For binary vectors (e.g., Hamming, Jaccard), this is the number of bits. In
+>                         case of SimSIMD, the inputs will be treated at the bit-level. Other packages will be matching/comparing 8-bit integers. The
+>                         volume of exchanged data will be identical, but the results will differ.
+>   -n COUNT, --count COUNT
+>                         Number of vectors per batch (default: 1) By default, when set to 1 the benchmark will generate many vectors of size (ndim, )
+>                         and call the functions on pairs of single vectors: both directly, and through `cdist`. Alternatively, for larger batch sizes
+>                         the benchmark will generate two matrices of size (n, ndim) and compute: - batch mode: (n) distances between vectors in
+>                         identical rows of the two matrices, - all-pairs mode: (n^2) distances between all pairs of vectors in the two matrices via
+>                         `cdist`.
+>   --metric {all,dot,spatial,binary,probability,sparse}
+>                         Distance metric to use, profiles everything by default
+>   --dtype {all,bits,int8,uint16,uint32,float16,float32,float64,bfloat16,complex32,complex64,complex128}
+>                         Defines numeric types to benchmark, profiles everything by default
+>   --scipy               Profile SciPy, must be installed
+>   --scikit              Profile scikit-learn, must be installed
+>   --torch               Profile PyTorch, must be installed
+>   --tf                  Profile TensorFlow, must be installed
+>   --jax                 Profile JAX, must be installed
 ```
 
 
@@ -155,6 +197,12 @@ npm install -g bun
 bun test
 ```
 
+## Swift
+
+```sh
+swift build && swift test -v
+```
+
 ## GoLang
 
 ```sh

package/simsimd/Cargo.lock

@@ -496,7 +496,7 @@ dependencies = [
 
 [[package]]
 name = "simsimd"
-version = "5.6.4"
+version = "5.9.6"
 dependencies = [
  "cc",
  "criterion",

package/simsimd/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "simsimd"
-description = "Fastest SIMD-Accelerated Vector Similarity Functions for x86 and Arm"
-version = "5.6.4"
+description = "Portable mixed-precision BLAS-like vector math library for x86 and ARM"
+version = "5.9.6"
 edition = "2021"
 license = "Apache-2.0"
 authors = ["Ash Vardanian <1983160+ashvardanian@users.noreply.github.com>"]
@@ -30,12 +30,12 @@ cc = "1.0.83"
 [[bench]]
 name = "cosine"
 harness = false
-path = "rust/benches/cosine.rs"
+path = "scripts/bench_cosine.rs"
 
 [[bench]]
 name = "sqeuclidean"
 harness = false
-path = "rust/benches/sqeuclidean.rs"
+path = "scripts/bench_sqeuclidean.rs"
 
 [profile.bench]
 opt-level = 3     # Corresponds to -O3

package/simsimd/MANIFEST.in

@@ -1,2 +1,3 @@
 include include/simsimd/*.h
-include include/*
+include include/*
+include VERSION

package/simsimd/Package.swift

@@ -9,8 +9,20 @@ let package = Package(
         .library(name: "SimSIMD", targets: ["SimSIMD"]),
     ],
     targets: [
-        .testTarget(name: "SimSIMDTests", dependencies: ["SimSIMD"], path: "swift/tests"),
-        .target(name: "SimSIMD", dependencies: ["CSimSIMD"], path: "swift/source"),
-        .systemLibrary(name: "CSimSIMD", path: "include")
+        .testTarget(name: "Test", dependencies: ["SimSIMD"], path: "swift", exclude:["SimSIMD.swift"]),
+        .target(name: "SimSIMD", dependencies: ["CSimSIMD"], path: "swift", exclude:["Test.swift"]),
+        .target(
+            name: "CSimSIMD",
+            path: "include/simsimd/", // Adjust the path to include your C source files
+            sources: ["../../c/lib.c"], // Include the source file here
+            publicHeadersPath: ".",
+            cSettings: [
+                .define("SIMSIMD_DYNAMIC_DISPATCH", to: "1"), // Define a C macro
+                .define("SIMSIMD_NATIVE_F16", to: "0"), // Define a C macro
+                .define("SIMSIMD_NATIVE_BF16", to: "0"), // Define a C macro
+                .headerSearchPath("include/"), // Specify header search paths
+                .unsafeFlags(["-Wall"]) // Use with caution: specify custom compiler flags
+            ]
+        ),
     ]
 )