usearch 0.18.4 → 0.18.5

package/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "usearch"
-version = "0.18.4"
+version = "0.18.5"
 authors = ["Ashot Vardanian <1983160+ashvardanian@users.noreply.github.com>"]
 description = "Smaller & Faster Single-File Vector Search Engine from Unum"
 edition = "2021"

package/conanfile.py

@@ -4,7 +4,7 @@ from conans import ConanFile
 class USearchConan(ConanFile):
 
     name = 'USearch'
-    version = '0.18.4'
+    version = '0.18.5'
     license = 'Apache 2.0'
     url = 'https://github.com/unum-cloud/usearch'
     description = 'Smaller & Faster Single-File Vector Search Engine from Unum'

package/include/usearch/index_punned_dense.hpp

@@ -328,7 +328,7 @@ class index_punned_dense_gt {
         std::size_t vector_bytes = dimensions_ * sizeof(scalar_at);
 
         byte_t* casted_data = cast_buffer_.data() + casted_vector_bytes_ * config.thread;
-        bool casted = cast(vector_data, vector_bytes, casted_data);
+        bool casted = cast(vector_data, dimensions_, casted_data);
         if (casted)
             vector_data = casted_data, vector_bytes = casted_vector_bytes_, config.store_vector = true;
 
@@ -349,7 +349,7 @@ class index_punned_dense_gt {
         std::size_t vector_bytes = dimensions_ * sizeof(scalar_at);
 
         byte_t* casted_data = cast_buffer_.data() + casted_vector_bytes_ * config.thread;
-        bool casted = cast(vector_data, vector_bytes, casted_data);
+        bool casted = cast(vector_data, dimensions_, casted_data);
         if (casted)
             vector_data = casted_data, vector_bytes = casted_vector_bytes_;
 
@@ -365,7 +365,7 @@ class index_punned_dense_gt {
         std::size_t vector_bytes = dimensions_ * sizeof(scalar_at);
 
         byte_t* casted_data = cast_buffer_.data() + casted_vector_bytes_ * config.thread;
-        bool casted = cast(vector_data, vector_bytes, casted_data);
+        bool casted = cast(vector_data, dimensions_, casted_data);
         if (casted)
             vector_data = casted_data, vector_bytes = casted_vector_bytes_;
 
@@ -388,10 +388,10 @@ class index_punned_dense_gt {
         }
         member_citerator_t iterator = typed_->cbegin() + id;
         member_cref_t member = *iterator;
-        byte_t const* casted_vector = reinterpret_cast<byte_t const*>(member.vector.data());
-        bool casted = cast(casted_vector, casted_vector_bytes_, (byte_t*)reconstructed);
+        byte_t const* punned_vector = reinterpret_cast<byte_t const*>(member.vector.data());
+        bool casted = cast(punned_vector, dimensions_, (byte_t*)reconstructed);
         if (!casted)
-            std::memcpy(reconstructed, casted_vector, casted_vector_bytes_);
+            std::memcpy(reconstructed, punned_vector, casted_vector_bytes_);
         return true;
     }
 

package/include/usearch/index_punned_helpers.hpp

@@ -492,12 +492,11 @@ using memory_mapping_allocator_t = aligned_allocator_t;
 #endif
 
 template <typename from_scalar_at, typename to_scalar_at> struct cast_gt {
-    inline bool operator()(byte_t const* input, std::size_t bytes_in_input, byte_t* output) const {
+    inline bool operator()(byte_t const* input, std::size_t dimensions, byte_t* output) const {
         from_scalar_at const* typed_input = reinterpret_cast<from_scalar_at const*>(input);
         to_scalar_at* typed_output = reinterpret_cast<to_scalar_at*>(output);
-        std::size_t size_input = bytes_in_input / sizeof(from_scalar_at);
         auto converter = [](from_scalar_at from) { return to_scalar_at(from); };
-        std::transform(typed_input, typed_input + size_input, typed_output, converter);
+        std::transform(typed_input, typed_input + dimensions, typed_output, converter);
         return true;
     }
 };
@@ -523,23 +522,21 @@ template <> struct cast_gt<b1x8_t, b1x8_t> {
 };
 
 template <typename from_scalar_at> struct cast_gt<from_scalar_at, b1x8_t> {
-    inline bool operator()(byte_t const* input, std::size_t bytes_in_input, byte_t* output) const {
+    inline bool operator()(byte_t const* input, std::size_t dimensions, byte_t* output) const {
         from_scalar_at const* typed_input = reinterpret_cast<from_scalar_at const*>(input);
         unsigned char* typed_output = reinterpret_cast<unsigned char*>(output);
-        std::size_t size_input = bytes_in_input / sizeof(from_scalar_at);
-        for (std::size_t i = 0; i != size_input; ++i)
-            typed_output[i / CHAR_BIT] |= bool(typed_input[i]) ? (1 << (i & (CHAR_BIT - 1))) : 0;
+        for (std::size_t i = 0; i != dimensions; ++i)
+            typed_output[i / CHAR_BIT] |= bool(typed_input[i]) ? (128 >> (i & (CHAR_BIT - 1))) : 0;
         return true;
     }
 };
 
 template <typename to_scalar_at> struct cast_gt<b1x8_t, to_scalar_at> {
-    inline bool operator()(byte_t const* input, std::size_t bytes_in_input, byte_t* output) const {
+    inline bool operator()(byte_t const* input, std::size_t dimensions, byte_t* output) const {
         unsigned char const* typed_input = reinterpret_cast<unsigned char const*>(input);
         to_scalar_at* typed_output = reinterpret_cast<to_scalar_at*>(output);
-        std::size_t size_input = bytes_in_input * CHAR_BIT;
-        for (std::size_t i = 0; i != size_input; ++i)
-            typed_output[i] = bool(typed_input[i / CHAR_BIT] & (1 << (i & (CHAR_BIT - 1))));
+        for (std::size_t i = 0; i != dimensions; ++i)
+            typed_output[i] = bool(typed_input[i / CHAR_BIT] & (128 >> (i & (CHAR_BIT - 1))));
         return true;
     }
 };

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "usearch",
-    "version": "0.18.4",
+    "version": "0.18.5",
     "description": "Smaller & Faster Single-File Vector Search Engine from Unum",
     "author": "Ashot Vardanian",
     "license": "Apache 2.0",

package/python/lib.cpp

@@ -394,7 +394,8 @@ template <typename index_at> void view_index(index_at& index, std::string const&
 
 template <typename internal_at, typename external_at = internal_at, typename index_at = void>
 py::object get_typed_member(index_at const& index, label_t label) {
-    py::array_t<external_at> result_py(static_cast<Py_ssize_t>(index.scalar_words()));
+    std::size_t result_slots = std::is_same<internal_at, b1x8_t>() ? index.scalar_words() : index.dimensions();
+    py::array_t<external_at> result_py(static_cast<Py_ssize_t>(result_slots));
     auto result_py1d = result_py.template mutable_unchecked<1>();
     if (!index.get(label, (internal_at*)&result_py1d(0)))
         return py::none();

package/python/scripts/bench.py

@@ -4,7 +4,6 @@ import itertools
 from typing import List, Optional, Iterable
 from dataclasses import asdict
 
-import fire
 import numpy as np
 import pandas as pd
 

package/python/scripts/test.py

@@ -5,7 +5,7 @@ import numpy as np
 from usearch.io import load_matrix, save_matrix
 from usearch.eval import recall_members, random_vectors
 
-from usearch.index import Index, SparseIndex, MetricKind, ScalarKind, Matches
+from usearch.index import Index, SparseIndex, MetricKind, ScalarKind, Matches, MetricKindBitwise
 from usearch.index import _normalize_dtype
 from usearch.index import (
     DEFAULT_CONNECTIVITY,
@@ -181,10 +181,13 @@ def test_bitwise_index(bits: int, metric: MetricKind, connectivity: int, batch_s
     index = Index(ndim=bits, metric=metric, connectivity=connectivity)
 
     labels = np.arange(batch_size)
-    bit_vectors = np.random.randint(2, size=(batch_size, bits))
-    bit_vectors = np.packbits(bit_vectors, axis=1)
+    byte_vectors = np.random.randint(2, size=(batch_size, bits))
+    bit_vectors = np.packbits(byte_vectors, axis=1)
 
     index.add(labels, bit_vectors)
+    assert np.allclose(index.get_vectors(labels), byte_vectors, atol=0.1)
+    assert np.alltrue(index.get_vectors(labels, ScalarKind.B1) == bit_vectors)
+    
     index.search(bit_vectors, 10)