• Stars
    star
    691
  • Rank 65,435 (Top 2 %)
  • Language
    Python
  • Created almost 7 years ago
  • Updated over 1 year ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Exploring word2vec embeddings as a graph of nearest neighbors

word2vec graph

This visualization builds graphs of nearest neighbors from high-dimensional word2vec embeddings.

demo 1 demo 2 demo words

Available Graphs

The dataset used for this visualization comes from GloVe, and has 6B tokens, 400K vocabulary, 300-dimensional vectors.

  • Distance < 0.9 - In this visualization edge between words is formed when distance between corresponding words' vectors is smaller than 0.9. All words with non-word characters and digits are removed. The final visualization is sparse, yet meaningful.

  • Distance < 1.0 - Similar to above, yet distance requirement is relaxed. Words with distance smaller than 1.0 are given edges in the graph. All words with non-word characters and digits are removed. The visualization becomes more populated as more words are added. Still meaningful.

  • Raw; Distance < 0.9 (6.9 MB) - Unlike visualizations above, this one was not filtered and includes all words from the dataset. Majority of the clusters formed here have numerical nature. I didn't find this one particularly interesting, yet I'm including it to show how word2vec finds numerical clusters.

Common Crawl

I have also made a graph from Common Crawl dataset (840B tokens, 2.2M vocab, 300d vectors). Words with non-word characters and numbers were removed.

Many clusters that remained represent words with spelling errors:

spelling error

I had hard time deciphering meaning of many clusters here. Wikipedia embeddings were much more meaningful. Nevertheless I want to keep this visualization to let you explore it as well:

Intro and Details

word2vec is a family of algorithms that allow you to find embeddings of words into high-dimensional vector spaces.

// For example
cat => [0.1, 0.0, 0.9]
dog => [0.9, 0.0, 0.0]
cow => [0.6, 1.0, 0.5]

Vectors with shorter distances between them usually share common contexts in the corpus. This allows us to find distances between words:

|cat - dog| = 1.20
|cat - cow| = 1.48


"cat" is closer to "dog" than it is to the "cow".

Building a graph

We can simply iterate over every single word in the dictionary and add them into a graph. But what would be an edge in this graph?

We draw an edge between two words if distance between embedding vectors is shorter than a given threshold.

Once the graph is constructed, I'm using a method described here: Your own graphs to construct visualizations.

Note From practical standpoint, searching all nearest neighbors in high dimensional space is a very CPU intensive task. Building an index of vectors help. I didn't know a good library for this task, so I consulted Twitter. Amazing recommendations by @gumgumeo and @AMZoellner led to spotify/annoy.

Data

I'm using pre-trained word2vec models from the GloVe project.

Preprocessing

My original attempts to render word2vec graphs resulted in overwhelming presence of numerical clusters. word2vec models really loved to put numerals together (and I think it makes sense, intuitively). Alas - that made visualizations not very interesting to explore. As I hoped from one cluster to another, just to find out that one was dedicated to numbers 2017 - 2300, while the other to 0.501 .. 0.403

In Common Crawl word2vec encoding, I removed all words that had non-word characters or numbers. In my opinion, this made visualization more interesting to explore, yet still, I don't recognize a lot of clusters.

Local setup

Prerequisites

Make sure node.js is installed.

git clone https://github.com/anvaka/word2vec-graph.git
cd word2vec-graph
npm install

Install spotify/annoy

Building graph file

  1. Download the vectors, and extract them into graph-data
  2. Run save_text_edges.py -h to see how to point it to th newly extracted. vectors (also see file content for more details)
  3. run python save_text_edges.py - depending on input vector file size this make take a while. The output file edges.txt will be saved in the graph-data folder.
  4. run node edges2graph.js graph-data/edges.txt - this will save graph in binary format into graph-data folder (graph-data/labels.json, graph-data/links.bin)
  5. Now it's time to run layout. There are two options. One is slow, the other one is much faster especially on the multi-threaded CPU.

Running layout with node

You can use

node --max-old-space-size=12000 layout.js

To generate layout. This will take a while to converge (layout stops after 500 iterations). Also note, that we need to increase maximum allowed RAM for node process (max-old-space-size argument). I'm setting it to ~12GB - it was enough for my case

Running layout with C++

Much faster version is to compile layout++ module. You will need to manually download and compile anvaka/ngraph.native package.

On ubuntu it was very straightforward: Just run ./compile-demo and layout++ file will be created in the working folder. You can copy that file into this repository, and run:

./layout++ ./graph-data/links.bin

The layout will converge much faster, but you will need to manually kill it (Ctrl + C) after 500-700 iterations.

You will find many .bin files. Just pick the one with the highest number, and copy it as positions.bin into graph-data/ folder. E.g.:

cp 500.bin ./graph-data/positions.bin

That's it. Now you have both graph, and positions ready. You can use instructions from Your own graphs to visualize your new graph with https://anvaka.github.io/pm/#/

More Repositories

1

city-roads

Visualization of all roads within any city
JavaScript
5,402
star
2

VivaGraphJS

Graph drawing library for JavaScript
JavaScript
3,646
star
3

ngraph.path

Path finding in a graph
JavaScript
2,801
star
4

atree

Just a simple Christmas tree, based on reddit story
JavaScript
2,424
star
5

panzoom

Universal pan and zoom library (DOM, SVG, Custom)
JavaScript
1,570
star
6

pm

package managers visualization
JavaScript
1,408
star
7

ngraph

Beautiful Graphs
1,360
star
8

npmgraph.an

2d visualization of npm
JavaScript
1,160
star
9

fieldplay

A vector field explorer
JavaScript
1,108
star
10

sayit

Visualization of related subreddits
JavaScript
937
star
11

vs

Visualization of Google's autocomplete
JavaScript
919
star
12

time

Simple Google Sheets interface to track time
JavaScript
623
star
13

graph-drawing-libraries

Trying to compare known graph drawing libraries
JavaScript
584
star
14

peak-map

Make a ridge line chart from any region on Earth
JavaScript
583
star
15

map-of-reddit

Interactive map of reddit
JavaScript
540
star
16

common-words

visualization of common words in different programming languages
JavaScript
495
star
17

ngraph.graph

Graph data structure in JavaScript
JavaScript
463
star
18

ngraph.pixel

Fast graph renderer based on low level ShaderMaterial from three.js
JavaScript
295
star
19

npmrank

npm dependencies graph metrics
JavaScript
284
star
20

streamlines

Streamlines calculator
JavaScript
275
star
21

isect

Segments intersection detection library
JavaScript
253
star
22

gauss-distribution

A fun little project to show distribution of pixels in Gauss's portrait
HTML
249
star
23

oflow

Optical flow detection in JavaScript
JavaScript
200
star
24

ghindex

Creates github index for similar repositories discovery
JavaScript
189
star
25

gazer

GitHub analysis and discovery
JavaScript
186
star
26

git-also

For a `file` in your git repository, prints other files that are most often committed together
JavaScript
185
star
27

allnpmviz3d

3d visualization of npm
JavaScript
179
star
28

ngraph.path.demo

This is a demo project for ngraph.path
JavaScript
165
star
29

map-of-github

Inspirational Mapping
Vue
157
star
30

ngraph.forcelayout

Force directed graph layout
JavaScript
146
star
31

pixchart

Turn any image into delightful splash of colors and order
JavaScript
112
star
32

city-script

Collection of scripts that can be loaded into city-roads
JavaScript
112
star
33

winvelviz

Wind visualization over time
JavaScript
100
star
34

yasiv-youtube

Graph of related videos from YouTube
JavaScript
100
star
35

query-state

Application state in query string
JavaScript
97
star
36

e-sum

Visualization of exponential sums
JavaScript
97
star
37

ngraph.forcelayout3d

Force directed graph layout in 3d
JavaScript
95
star
38

index-large-cities

A simple indexer of road networks from OSM. Data for @anvaka/city-roads
JavaScript
92
star
39

graph-start

a simple graph shell to explore ideas
JavaScript
88
star
40

greview

Books that I read and their neighborhoods
86
star
41

lsystem

A simple L-Systems explorer powered by WebGL
JavaScript
86
star
42

jsruntime

Chrome Extension to explore javascript runtime.
JavaScript
85
star
43

map-of-reddit-data

Contains scripts and data to render map of reddit
JavaScript
82
star
44

redsim

reddit discovery
JavaScript
82
star
45

w-gl

A simple WebGL renderer
TypeScript
80
star
46

pplay

Create, play and share pixels. Online WebGL shader editor.
GLSL
76
star
47

ngraph.hde

High dimensional embedding of a graph and its layout
JavaScript
76
star
48

dotparser

Parser of GraphViz dot file format
PEG.js
72
star
49

wind-lines

Streamline animation of wind data
JavaScript
63
star
50

set-vs-object

What is faster Set or Object?
JavaScript
62
star
51

three.map.control

A three.js camera that mimics 2d maps navigation with pan and zoom
JavaScript
53
star
52

ngraph.native

C++ implementation of force-based layout from ngraph
C++
49
star
53

circles

A simple spirograph toy
JavaScript
49
star
54

yaot

Yet another octree
JavaScript
48
star
55

ngraph.three

3D graph renderer powered by three.js
JavaScript
44
star
56

ngraph.generators

Graph generators
JavaScript
42
star
57

citations

Most cited papers by keyword
C++
42
star
58

rafor

requestAnimationFrame friendly async for iterator
JavaScript
41
star
59

ngraph.fabric

Fabric.js graph renderer
JavaScript
38
star
60

playground

Just a set of experiments that I want to play with, but they are too small to be in their own repository
JavaScript
35
star
61

ngraph.centrality

Module to calculate graph centrality metrics
JavaScript
33
star
62

streak

Streak tracking with Google Sheets
JavaScript
33
star
63

sayit-data

data with similar subreddits graph
JavaScript
32
star
64

ngraph.offline.layout

Performs offline layout of large graphs and saves results to the disk
JavaScript
32
star
65

cord-19

exploring research papers about coronaviruses
JavaScript
31
star
66

how-to-debug-node-js-addons

How to debug node.js addons in xcode
30
star
67

wheel

Mouse wheel event unified for all browsers
JavaScript
30
star
68

npmgraphbuilder

Builds graph of npm dependencies from npm registry
JavaScript
29
star
69

generator-n

minimalistic node package yeoman generator
JavaScript
28
star
70

allgithub

Crawling github data
JavaScript
27
star
71

tiny.xml

Tiny (1.6KB) in-browser xml parser
JavaScript
27
star
72

mars

Map of Mars
JavaScript
26
star
73

ngraph.pixi

PIXI.js graph renderer
JavaScript
26
star
74

ngraph.pagerank

PageRank calculation for ngraph.graph
JavaScript
25
star
75

noisylines

Tracking noise with streamlines
JavaScript
23
star
76

nb

Neighborhood beautification: Graph layout through message passing
JavaScript
23
star
77

ngraph.physics.simulator

Physics library for ngraph
JavaScript
23
star
78

strangeb

The strangest thing happens when you rotate Bezier control points
JavaScript
22
star
79

graph-to-vector-field

Converts a graph into vector field texture
JavaScript
20
star
80

amator

Tiny animation library
JavaScript
20
star
81

ngraph.louvain

Given a graph instance detects communities using the Louvain Method
JavaScript
20
star
82

npmgraph

Visualization of NPM dependencies
JavaScript
19
star
83

similar-cities

Visualization of cities with similar road networks
JavaScript
19
star
84

allnpm

Graph generator for entire npm registry
JavaScript
18
star
85

twitter-recommended-graph

Building a proposal for Twitter to show a map of recommended people
JavaScript
18
star
86

streaming-svg-parser

Streaming SVG/XML parser with zero dependencies
JavaScript
18
star
87

extract-osm-roads

A simple utility to fetch a city graph from OSM
JavaScript
18
star
88

quadtree.cc

A C++ implementation of quadtree
C++
17
star
89

rules-of-ml

A simple visualization of Martin Zinkevich article
JavaScript
17
star
90

ngraph.events

Events support in ngraph.*
JavaScript
17
star
91

simplegrad

Simple reverse mode automatic differentiation of scalar values in javascript
JavaScript
16
star
92

portrait

Portrait of quotes
JavaScript
16
star
93

sunburst

For a given tree builds an SVG based SunBurst diagram
JavaScript
15
star
94

local-chat

Local instance of ChatGPT for my kiddo
HTML
15
star
95

what-people-google

Visualization of what people google
JavaScript
15
star
96

vuereddit

A simple reddit client written as a vue component.
Vue
14
star
97

ngraph.fromjson

Library to load graph from simple json format
JavaScript
12
star
98

color-high

A demo of ngraph.forcelayout in 6D space
JavaScript
12
star
99

color-force-vis

Visualizing forces acting on nodes during force layout
JavaScript
11
star
100

allnpmviz.an

Visualization of entire npm
JavaScript
11
star