This repository is for MN-RU algorithm. All experiment has implemented in C++ on a PC with Intel Xeon CPU E5-2678 v3 @ 2.50GHz and 110GB memory, running Ubuntu 22.04.

You could reproduce it with the following steps, the overall experiment may take days.

Suppose we are in the root directory of the repo.

before start, you need to:

apt update && \
apt install -y g++ && \
apt install -y cmake && \
apt install -y build-essential && \
apt install -y make && \
apt install wget && \
apt install zip unzip && \
apt install tar && \
apt install python3 && \
apt install python3-dev python3-pip && \
pip install numpy pandas matplotlib

evaluation

Our repository contains the results in the ./output directory and its subdirectories. You can view them directly or run the following command to see the results.

1. download datasets

bash ./shell/downloadDatasets.sh

2. build the project

bash ./shell/build.sh

3. special deal with sift2M

bash ./shell/specialDealWithSift2M.sh

4. generate index and groundTruth

bash ./shell/generateIndexAndGroundTruth.sh

5. run full_coverage scenario

bash ./shell/runMultWithBuild.sh

6. run random scenario

bash ./shell/runRandomWithBuild.sh

7. run new_insert scenario

bash ./shell/runNewInsertWithBuild.sh

8. run backup

bash ./shell/runBackUp.sh

9. Draw the result Figure

bash ./shell/runDrawFigure.sh

Viewing Results

After drawing the result figures, you can find them in the ./output directory and its subdirectories.

MN-RU algorithm position

• MN-RU α: ./external/hnswlib_method7/hnswlib/hnswalg.h
• MN-RU β: ./external/hnswlib_method8/hnswlib/hnswalg.h
• MN-RU γ: ./external/hnswlib_method9/hnswlib/hnswalg.h
• MN-THN-RU: ./external/hnswlib_method10/hnswlib/hnswalg.h You could just see the core algorithm in function updatePoint and repairConnectionsForUpdate