Comments (3)
does n2 handles duplicate vectors without performance degradation or recall issues?
Yes, since version 0.1.7
The HNSW algorithm doesn't work efficiently on duplicate vectors. We thought this was because the heuristic neighbor selection algorithm focused only on navigation. With the heuristic neighbor selection, duplicate or near-duplicate vectors are hidden and
search becomes difficult, resulting in a low recall.
To solve this, we modified the heuristic neighbor selection algorithm and improved it in a form that has some nearest neighbors but does not degrade navigation performance.
Below is one of the benchmarks measured for the 0.1.7 release, and GIST has duplicate vectors(about 2% of train vectors are duplicated)
You can see a high recall compared to N2 version 0.1.6
Handling duplicate vectors have a tradeoff relationship with navigation performance, the way we handled it may not be optimal. So we are continuing to work to find if there is a better way.
from n2.
It's awesome that you're tackling this problem, thank you very much for the detailed response. Please feel free to close the issue if you wish.
from n2.
If we found another achievement for that problem, I'll comment here.
from n2.
Related Issues (20)
- Windows installation issue HOT 1
- With the same accuracy, the number of comparisons with the paper is very different. HOT 3
- The core idea behind this library
- Loading model with mmap lowering searching speed? HOT 4
- Does it support online vector deletion and insertion? HOT 1
- Can't install into Conda environment on macOS Big Sur HOT 1
- How dose n2 handle zero vectors? HOT 3
- multiple indices HOT 1
- Support Hamming-Distance
- Cannot use with pytorch due to libgomp issue HOT 1
- missing xmmintrin.h on ARM devices. HOT 1
- [Go binding] Linux make error: make: *** [go] Error 1
- Custom Metric
- N2 in Big-ANN-Benchmarks NeurIPS'21 Challenge?
- `Library not loaded: /usr/local/opt/gcc/lib/gcc/10/libgomp.1.dylib` HOT 1
- What are the right parameter ?
- extended candidates and keep pruned
- Windows installation HOT 1
- Can we add support for PPC big endian?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from n2.