Skip to content

High-speed approximate nearest neighbors for Ruby

License

Notifications You must be signed in to change notification settings

ankane/ngt-ruby

Repository files navigation

NGT Ruby

NGT - high-speed approximate nearest neighbors - for Ruby

Build Status

Installation

Add this line to your application’s Gemfile:

gem "ngt"

On Mac, also install OpenMP:

brew install libomp

NGT is not available for Windows

Getting Started

Prep your data

objects = [
  [1, 1, 2, 1],
  [5, 4, 6, 5],
  [1, 2, 1, 2]
]

Create an index

index = Ngt::Index.new(dimensions)

Insert objects

index.batch_insert(objects)

Search the index

index.search(query, size: 3)

Save the index

index.save(path)

Load an index

index = Ngt::Index.load(path)

Get an object by id

index.object(id)

Insert a single object

index.insert(object)

Remove an object by id

index.remove(id)

Build the index

index.build_index

Optimize the index

optimizer = Ngt::Optimizer.new(outgoing: 10, incoming: 120)
optimizer.adjust_search_coefficients(index)
optimizer.execute(index, new_path)

Full Example

dim = 10
objects = []
100.times do |i|
  objects << dim.times.map { rand(100) }
end

index = Ngt::Index.new(dim)
index.batch_insert(objects)

query = objects[0]
result = index.search(query, size: 3)

result.each do |res|
  puts "#{res[:id]}, #{res[:distance]}"
  p index.object(res[:id])
end

Index Options

Defaults shown below

Ngt::Index.new(dimensions,
  edge_size_for_creation: 10,
  edge_size_for_search: 40,
  object_type: :float, # :float, :integer
  distance_type: :l2,  # :l1, :l2, :hamming, :angle, :cosine, :normalized_angle, :normalized_cosine, :jaccard
  path: nil
)

Optimizer Options

Defaults shown below

Ngt::Optimizer.new(
  outgoing: 10,
  incoming: 120,
  queries: 100,
  low_accuracy_from: 0.3,
  low_accuracy_to: 0.5,
  high_accuracy_from: 0.8,
  high_accuracy_to: 0.9,
  gt_epsilon: 0.1,
  merge: 0.2
)

Data

Data can be an array of arrays

[[1, 2, 3], [4, 5, 6]]

Or a Numo array

Numo::NArray.cast([[1, 2, 3], [4, 5, 6]])

Resources

Credits

This library is modeled after NGT’s Python API.

History

View the changelog

Contributing

Everyone is encouraged to help improve this project. Here are a few ways you can help:

To get started with development:

git clone https://github.com/ankane/ngt-ruby.git
cd ngt-ruby
bundle install
bundle exec rake vendor:all
bundle exec rake test

About

High-speed approximate nearest neighbors for Ruby

Resources

License

Stars

Watchers

Forks

Packages

No packages published