Add cpp examples (#435)

* Add cpp examples
* Add multithreaded cpp examples

.github/workflows/build.yml

@@ -21,7 +21,7 @@ jobs:
       - name: Test
         timeout-minutes: 15
         run: |
-          python -m unittest discover -v --start-directory examples --pattern "example*.py"
+          python -m unittest discover -v --start-directory examples/python --pattern "example*.py"
           python -m unittest discover -v --start-directory tests/python --pattern "bindings_test*.py"
   
   test_cpp:
@@ -61,6 +61,12 @@ jobs:
           if [ "$RUNNER_OS" == "Windows" ]; then
             cp ./Release/* ./
           fi
+          ./example_search
+          ./example_filter
+          ./example_replace_deleted
+          ./example_mt_search
+          ./example_mt_filter
+          ./example_mt_replace_deleted
           ./searchKnnCloserFirst_test
           ./searchKnnWithFilter_test
           ./multiThreadLoad_test

ALGO_PARAMS.md

@@ -27,5 +27,5 @@ ef_construction leads to longer construction, but better index quality. At some
 not improve the quality of the index. One way to check if the selection of ef_construction was ok is to measure a recall 
 for M nearest neighbor search when ```ef``` =```ef_construction```: if the recall is lower than 0.9, than there is room 
 for improvement.
-* ```num_elements``` - defines the maximum number of elements in the index. The index can be extened by saving/loading(load_index
+* ```num_elements``` - defines the maximum number of elements in the index. The index can be extended by saving/loading (load_index
 function has a parameter which defines the new maximum number of elements).

CMakeLists.txt

@@ -16,6 +16,26 @@ if(CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
       SET( CMAKE_CXX_FLAGS  "-Ofast -lrt -DNDEBUG -std=c++11 -DHAVE_CXX0X -openmp -march=native -fpic -w -fopenmp -ftree-vectorize" )
     endif()
 
+    # examples
+    add_executable(example_search examples/cpp/example_search.cpp)
+    target_link_libraries(example_search hnswlib)
+
+    add_executable(example_filter examples/cpp/example_filter.cpp)
+    target_link_libraries(example_filter hnswlib)
+
+    add_executable(example_replace_deleted examples/cpp/example_replace_deleted.cpp)
+    target_link_libraries(example_replace_deleted hnswlib)
+
+    add_executable(example_mt_search examples/cpp/example_mt_search.cpp)
+    target_link_libraries(example_mt_search hnswlib)
+
+    add_executable(example_mt_filter examples/cpp/example_mt_filter.cpp)
+    target_link_libraries(example_mt_filter hnswlib)
+
+    add_executable(example_mt_replace_deleted examples/cpp/example_mt_replace_deleted.cpp)
+    target_link_libraries(example_mt_replace_deleted hnswlib)
+
+    # tests
     add_executable(test_updates tests/cpp/updates_test.cpp)
     target_link_libraries(test_updates hnswlib)
 

README.md

@@ -111,7 +111,12 @@ Properties of `hnswlib.Index` that support reading and writing:
         
 
 #### Python bindings examples
-[See more examples here](examples/EXAMPLES.md)
+[See more examples here](examples/python/EXAMPLES.md):
+* Creating index, inserting elements, searching, serialization/deserialization
+* Filtering during the search with a boolean function
+* Deleting the elements and reusing the memory of the deleted elements for newly added elements
+
+An example of creating index, inserting elements, searching and pickle serialization:
 ```python
 import hnswlib
 import numpy as np
@@ -218,6 +223,14 @@ labels, distances = p.knn_query(data, k=1)
 print("Recall for two batches:", np.mean(labels.reshape(-1) == np.arange(len(data))), "\n")
 ```
 
+#### C++ examples
+[See examples here](examples/cpp/EXAMPLES.md):
+* creating index, inserting elements, searching, serialization/deserialization
+* filtering during the search with a boolean function
+* deleting the elements and reusing the memory of the deleted elements for newly added elements
+* multithreaded usage
+
+
 ### Bindings installation
 
 You can install from sources:

examples/cpp/EXAMPLES.md

@@ -0,0 +1,185 @@
+# C++ examples
+
+Creating index, inserting elements, searching and serialization
+```cpp
+#include "../../hnswlib/hnswlib.h"
+
+
+int main() {
+    int dim = 16;               // Dimension of the elements
+    int max_elements = 10000;   // Maximum number of elements, should be known beforehand
+    int M = 16;                 // Tightly connected with internal dimensionality of the data
+                                // strongly affects the memory consumption
+    int ef_construction = 200;  // Controls index search speed/build speed tradeoff
+
+    // Initing index
+    hnswlib::L2Space space(dim);
+    hnswlib::HierarchicalNSW<float>* alg_hnsw = new hnswlib::HierarchicalNSW<float>(&space, max_elements, M, ef_construction);
+
+    // Generate random data
+    std::mt19937 rng;
+    rng.seed(47);
+    std::uniform_real_distribution<> distrib_real;
+    float* data = new float[dim * max_elements];
+    for (int i = 0; i < dim * max_elements; i++) {
+        data[i] = distrib_real(rng);
+    }
+
+    // Add data to index
+    for (int i = 0; i < max_elements; i++) {
+        alg_hnsw->addPoint(data + i * dim, i);
+    }
+
+    // Query the elements for themselves and measure recall
+    float correct = 0;
+    for (int i = 0; i < max_elements; i++) {
+        std::priority_queue<std::pair<float, hnswlib::labeltype>> result = alg_hnsw->searchKnn(data + i * dim, 1);
+        hnswlib::labeltype label = result.top().second;
+        if (label == i) correct++;
+    }
+    float recall = correct / max_elements;
+    std::cout << "Recall: " << recall << "\n";
+
+    // Serialize index
+    std::string hnsw_path = "hnsw.bin";
+    alg_hnsw->saveIndex(hnsw_path);
+    delete alg_hnsw;
+
+    // Deserialize index and check recall
+    alg_hnsw = new hnswlib::HierarchicalNSW<float>(&space, hnsw_path);
+    correct = 0;
+    for (int i = 0; i < max_elements; i++) {
+        std::priority_queue<std::pair<float, hnswlib::labeltype>> result = alg_hnsw->searchKnn(data + i * dim, 1);
+        hnswlib::labeltype label = result.top().second;
+        if (label == i) correct++;
+    }
+    recall = (float)correct / max_elements;
+    std::cout << "Recall of deserialized index: " << recall << "\n";
+
+    delete[] data;
+    delete alg_hnsw;
+    return 0;
+}
+```
+
+An example of filtering with a boolean function during the search:
+```cpp
+#include "../../hnswlib/hnswlib.h"
+
+
+// Filter that allows labels divisible by divisor
+class PickDivisibleIds: public hnswlib::BaseFilterFunctor {
+unsigned int divisor = 1;
+ public:
+    PickDivisibleIds(unsigned int divisor): divisor(divisor) {
+        assert(divisor != 0);
+    }
+    bool operator()(hnswlib::labeltype label_id) {
+        return label_id % divisor == 0;
+    }
+};
+
+
+int main() {
+    int dim = 16;               // Dimension of the elements
+    int max_elements = 10000;   // Maximum number of elements, should be known beforehand
+    int M = 16;                 // Tightly connected with internal dimensionality of the data
+                                // strongly affects the memory consumption
+    int ef_construction = 200;  // Controls index search speed/build speed tradeoff
+
+    // Initing index
+    hnswlib::L2Space space(dim);
+    hnswlib::HierarchicalNSW<float>* alg_hnsw = new hnswlib::HierarchicalNSW<float>(&space, max_elements, M, ef_construction);
+
+    // Generate random data
+    std::mt19937 rng;
+    rng.seed(47);
+    std::uniform_real_distribution<> distrib_real;
+    float* data = new float[dim * max_elements];
+    for (int i = 0; i < dim * max_elements; i++) {
+        data[i] = distrib_real(rng);
+    }
+
+    // Add data to index
+    for (int i = 0; i < max_elements; i++) {
+        alg_hnsw->addPoint(data + i * dim, i);
+    }
+
+    // Create filter that allows only even labels
+    PickDivisibleIds pickIdsDivisibleByTwo(2);
+
+    // Query the elements for themselves with filter and check returned labels
+    int k = 10;
+    for (int i = 0; i < max_elements; i++) {
+        std::vector<std::pair<float, hnswlib::labeltype>> result = alg_hnsw->searchKnnCloserFirst(data + i * dim, k, &pickIdsDivisibleByTwo);
+        for (auto item: result) {
+            if (item.second % 2 == 1) std::cout << "Error: found odd label\n";
+        }
+    }
+
+    delete[] data;
+    delete alg_hnsw;
+    return 0;
+}
+```
+
+An example with reusing the memory of the deleted elements when new elements are being added (via `allow_replace_deleted` flag):
+```cpp
+#include "../../hnswlib/hnswlib.h"
+
+
+int main() {
+    int dim = 16;               // Dimension of the elements
+    int max_elements = 10000;   // Maximum number of elements, should be known beforehand
+    int M = 16;                 // Tightly connected with internal dimensionality of the data
+                                // strongly affects the memory consumption
+    int ef_construction = 200;  // Controls index search speed/build speed tradeoff
+
+    // Initing index
+    hnswlib::L2Space space(dim);
+    hnswlib::HierarchicalNSW<float>* alg_hnsw = new hnswlib::HierarchicalNSW<float>(&space, max_elements, M, ef_construction, 100, true);
+
+    // Generate random data
+    std::mt19937 rng;
+    rng.seed(47);
+    std::uniform_real_distribution<> distrib_real;
+    float* data = new float[dim * max_elements];
+    for (int i = 0; i < dim * max_elements; i++) {
+        data[i] = distrib_real(rng);
+    }
+
+    // Add data to index
+    for (int i = 0; i < max_elements; i++) {
+        alg_hnsw->addPoint(data + i * dim, i);
+    }
+
+    // Mark first half of elements as deleted
+    int num_deleted = max_elements / 2;
+    for (int i = 0; i < num_deleted; i++) {
+        alg_hnsw->markDelete(i);
+    }
+
+    float* add_data = new float[dim * num_deleted];
+    for (int i = 0; i < dim * num_deleted; i++) {
+        add_data[i] = distrib_real(rng);
+    }
+
+    // Replace deleted data with new elements
+    // Maximum number of elements is reached therefore we cannot add new items,
+    // but we can replace the deleted ones by using replace_deleted=true
+    for (int i = 0; i < num_deleted; i++) {
+        int label = max_elements + i;
+        alg_hnsw->addPoint(add_data + i * dim, label, true);
+    }
+
+    delete[] data;
+    delete[] add_data;
+    delete alg_hnsw;
+    return 0;
+}
+```
+
+Multithreaded examples:
+* Creating index, inserting elements, searching [example_mt_search.cpp](example_mt_search.cpp)
+* Filtering during the search with a boolean function [example_mt_filter.cpp](example_mt_filter.cpp)
+* Reusing the memory of the deleted elements when new elements are being added [example_mt_replace_deleted.cpp](example_mt_replace_deleted.cpp)

examples/cpp/example_filter.cpp

@@ -0,0 +1,57 @@
+#include "../../hnswlib/hnswlib.h"
+
+
+// Filter that allows labels divisible by divisor
+class PickDivisibleIds: public hnswlib::BaseFilterFunctor {
+unsigned int divisor = 1;
+ public:
+    PickDivisibleIds(unsigned int divisor): divisor(divisor) {
+        assert(divisor != 0);
+    }
+    bool operator()(hnswlib::labeltype label_id) {
+        return label_id % divisor == 0;
+    }
+};
+
+
+int main() {
+    int dim = 16;               // Dimension of the elements
+    int max_elements = 10000;   // Maximum number of elements, should be known beforehand
+    int M = 16;                 // Tightly connected with internal dimensionality of the data
+                                // strongly affects the memory consumption
+    int ef_construction = 200;  // Controls index search speed/build speed tradeoff
+
+    // Initing index
+    hnswlib::L2Space space(dim);
+    hnswlib::HierarchicalNSW<float>* alg_hnsw = new hnswlib::HierarchicalNSW<float>(&space, max_elements, M, ef_construction);
+
+    // Generate random data
+    std::mt19937 rng;
+    rng.seed(47);
+    std::uniform_real_distribution<> distrib_real;
+    float* data = new float[dim * max_elements];
+    for (int i = 0; i < dim * max_elements; i++) {
+        data[i] = distrib_real(rng);
+    }
+
+    // Add data to index
+    for (int i = 0; i < max_elements; i++) {
+        alg_hnsw->addPoint(data + i * dim, i);
+    }
+
+    // Create filter that allows only even labels
+    PickDivisibleIds pickIdsDivisibleByTwo(2);
+
+    // Query the elements for themselves with filter and check returned labels
+    int k = 10;
+    for (int i = 0; i < max_elements; i++) {
+        std::vector<std::pair<float, hnswlib::labeltype>> result = alg_hnsw->searchKnnCloserFirst(data + i * dim, k, &pickIdsDivisibleByTwo);
+        for (auto item: result) {
+            if (item.second % 2 == 1) std::cout << "Error: found odd label\n";
+        }
+    }
+
+    delete[] data;
+    delete alg_hnsw;
+    return 0;
+}