C++ Example

[ashfaq92]

Hi!
In your "readme.md" and "examples" folder, examples are provided only with python bindings.
In https://github.com/jlmelville/rcpphnsw rough equivalent of the serialization/deserialization example from this hnswlib README file is provided in 'R' language.
But nowhere C++ example of serialization/deserialization equivalent to current README example is provided. I want to use hnswlib in my c++ project. I was wondering how to do the exact functionality in c++. It would be great if you translate examples given in python to c++.
Thank you!

[yurymalkov]

@ashfaq92

Thanks for noticing! I'll make an example.

[jlmelville]

The basics of creating/re-sizing/re-loading an index, as used in RcppHNSW are below. It doesn't really do anything, but it should compile and run without causing your computer to explode, which is 90% of the battle in C++. Hopefully you can work out the rest!

#include <memory>
#include "hnswlib.h"
void example()
  auto dim = 16;
  auto num_elements = 100000;
  auto init_num_elements = num_elements / 2;
  auto M = 16;
  auto ef_construction = 200;

  using dist_t = float;
  using Distance = hnswlib::L2Space;

  // create the index
  auto space = std::unique_ptr<Distance>(new Distance(dim));
  auto appr_alg = std::unique_ptr<hnswlib::HierarchicalNSW<dist_t>>(
    new hnswlib::HierarchicalNSW<dist_t>(space.get(), init_num_elements, M,
                                         ef_construction));

  // resize it
  appr_alg->resizeIndex(num_elements);

  // or save and reload it
  std::string path_to_index = "myindex.hnsw";
  appr_alg->saveIndex(path_to_index);
  appr_alg.reset(new hnswlib::HierarchicalNSW<dist_t>(space.get(), path_to_index));
}

[ashfaq92]

@jlmelville thanks for prompt response. Sorry for late reply due to "timezones :(". I have run your code and it works without error (as you said :). But it would be really nice if you update this code according to the readme file. I mean if it make HNSW graph with random numbers and finds nearest neighbor for any random query point. Then we may add this code in readme files so that future c++ beginners may embed that in their code (like me).

[Akshaysharma29]

With the addition to the above code starter code. In this code querying the embedding is added(with query result).
`auto dim = 512;
auto num_elements = 100000;
auto init_num_elements = num_elements / 2;
auto M = 16;
auto ef_construction = 200;

using dist_t = float;
using Distance = hnswlib::L2Space;

// create the index
auto space = std::unique_ptr<Distance>(new Distance(dim));
auto appr_alg = std::unique_ptr<hnswlib::HierarchicalNSW<dist_t>>(
        new hnswlib::HierarchicalNSW<dist_t>(space.get(), init_num_elements, M,
                                             ef_construction));

typedef size_t labeltype;
appr_alg->setEf(200);
auto ret=appr_alg->searchKnn(default_emb,1);//'1' is the nearest neg

std::vector<std::pair<dist_t, labeltype>> result;//result return after knn search


int f=0;
int id_label;
float mn=100000;
if(ret.size()==0)
{
    counter_value+=1;
    appr_alg->addPoint(default_emb,counter_value);
    LOGD("This is empty ret");
} else{
    LOGD("Non empty ret");
    while (!ret.empty()) {
        result.push_back(ret.top());
        ret.pop();
    }
    for(int i =0;i<result.size();i++)
    {
        LOGD("hnswlib result:(1)distance:) %f => (2)label:) %d",result[i].first,result[i].second);
        if(result[i].first<(0.5)&&mn>result[i].first)
        {
            f=1;
            mn=result[i].first;
            id_label=result[i].second;
        }
    }
    if(f==0){
        counter_value+=1;
        id_label=counter_value;
        appr_alg->addPoint(default_emb,counter_value);
        LOGD("New entry generated");
    }else{
        LOGD("Id we get:%d",id_label);
    }
}

result.clear();`