Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Locality Sensitive Hashing at Lyst
Search
Maciej Kula
July 24, 2015
Programming
0
1.2k
Locality Sensitive Hashing at Lyst
Description of the intuition behind locality sensitive hashing and its application at Lyst.
Maciej Kula
July 24, 2015
Tweet
Share
More Decks by Maciej Kula
See All by Maciej Kula
Implicit and Explicit Recommender Systems
maciejkula
0
2.4k
Binary Embeddings For Efficient Ranking
maciejkula
0
600
Rust for Python Native Extensions
maciejkula
0
450
Hybrid Recommender Systems at PyData Amsterdam 2016
maciejkula
5
2.5k
Recommendations under sparsity
maciejkula
1
330
Metadata Embeddings for User and Item Cold-start Recommendations
maciejkula
2
790
Other Decks in Programming
See All in Programming
2 週間で Twitter Bot を作ってみた
contour_gara
0
390
Tailwind CSSを本気でカスタマイズする方法
fsubal
13
5.3k
スキーマ駆動開発による品質とスピードの両立 - 私達は何故、スキーマを書くのか
kentaroutakeda
0
170
FigmaとPHPで作る1ミリたりとも表示崩れしない最強の帳票印刷ソリューション
ttskch
43
19k
Micro Frontends for Java Microservices - Devnexus 2024
mraible
PRO
0
490
"config" ってなんだ? / What is "config"?
okashoi
0
240
TYPO3 v13 – The road to LTS: What's new and new APIs
luisasofie_xoxo
0
200
Fragment Composition of GraphQL
quramy
7
990
Snowflakeで眠ったデータを起こそう!
estie
0
120
OpenAPIを中心に考えるAPI開発入門 / Introduction to API Development with a Focus on OpenAPI
seike460
PRO
2
170
Java 22 Overview
kishida
1
180
What We Can Learn From OSS
inouehi
0
420
Featured
See All Featured
Large-scale JavaScript Application Architecture
addyosmani
504
110k
For a Future-Friendly Web
brad_frost
172
9k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
34
8.9k
The Cult of Friendly URLs
andyhume
74
5.7k
ピンチをチャンスに:未来をつくるプロダクトロードマップ #pmconf2020
aki_iinuma
78
42k
KATA
mclloyd
15
12k
Building Effective Engineering Teams - LeadDev
addyosmani
28
1.8k
How to train your dragon (web standard)
notwaldorf
73
5.2k
What the flash - Photography Introduction
edds
64
11k
Scaling GitHub
holman
457
140k
Mobile First: as difficult as doing things right
swwweet
216
8.6k
Automating Front-end Workflow
addyosmani
1356
200k
Transcript
Speeding up search with locality sensitive hashing. by Maciej Kula
Hi, I’m Maciej Kula. @maciej_kula
We collect the world of fashion into a customisable shopping
experience.
Given a point, find other points close to it. Nearest
neighbour search… 4
None
At Lyst we use it for… 1.) Image Search 2.)
Recommendations 6
Convert image to points in space (vectors) & use nearest
neighbour search to get similar images. 1. Image Search (-0.3, 2.1, 0.5)
Super useful for deduplication & search.
Convert products and users to points in space & use
nearest neighbour search to get related products for the user. 2. Recommendations user = (-0.3, 2.1, 0.5) product = (5.2, 0.3, -0.5)
Great, but…
11 80 million We have images
12 9 million We have products
Exhaustive nearest neighbour search is too slow.
Locality sensitive hashing to the rescue! Use a hash table.
Pick a hash function that puts similar points in the same bucket. Only search within the bucket.
We use Random Projection Forests
Partition by splitting on random vectors
Partition by splitting on random vectors
Partition by splitting on random vectors
Partition by splitting on random vectors
Partition by splitting on random vectors
Points to note Keep splitting until the nodes are small
enough. Median splits give nicely balanced trees. Build a forest of trees.
Why do we need a forest? Some partitions split the
true neighbourhood of a point. Because partitions are random, other trees will not repeat the error. Build more trees to trade off query speed for precision.
LSH in Python annoy, Python wrapper for C++ code. LSHForest,
part of scikit-learn FLANN, an auto-tuning ANN index
But… LSHForest is slow. FLANN is a pain to deploy.
annoy is great, but can’t add points to an existing index.
So we wrote our own.
github.com/lyst/rpforest pip install rpforest
rpforest Quite fast. Allows adding new items to the index.
Does not require us to store points in memory.
We use it in conjunction with PostgreSQL Send the query
point to the ANN index. Get ANN row ids back Plug them into postgres for filtering Final scoring done in postgres using C extensions.
Side note: postgres is awesome. Arrays & custom functions in
C
Gives us a fast and reliable ANN service 100x speed-up
with 0.6 10-NN precision Allows us to serve real-time results All on top of a real database.
thank you @maciej_kula