hanny 0.1.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 76848de111777349352ddeb7aa269ad3694504661ac2fe1c45729ebcd0f27414
+  data.tar.gz: a911da20689134ebecdd01b88827a2d2cc65149b497103f32f528fa754590c15
+SHA512:
+  metadata.gz: 6c1e4fde8dc03f439454a476d16be7b48f384cb0adf24a17fce47e09387d4583e31ba66b3a9b7f6d9fb5ff6e386e9282fba68f05c9ff0cb74f88013e19d65b91
+  data.tar.gz: 30441f1aef6a05bc0a609d4b2176f4e04d7d0d9882579de8909aae9076dbdf68670eb4275da1a863dffa13454997597e83d229a1371f6a51bee0f58aa3740abd

data/.coveralls.yml

@@ -0,0 +1 @@
+service_name: travis-ci

data/.gitignore

@@ -0,0 +1,15 @@
+/.bundle/
+/.yardoc
+/_yardoc/
+/coverage/
+/doc/
+/pkg/
+/spec/reports/
+/tmp/
+
+# rspec failure tracking
+.rspec_status
+
+*.swp
+.DS_Store
+.ruby-version

data/.rspec

@@ -0,0 +1,3 @@
+--format documentation
+--color
+--require spec_helper

data/.rubocop.yml

@@ -0,0 +1,39 @@
+AllCops:
+  TargetRubyVersion: 2.1
+  DisplayCopNames: true
+  DisplayStyleGuide: true
+
+Documentation:
+  Enabled: false
+
+Metrics/LineLength:
+  Max: 140
+  IgnoredPatterns: ['(\A|\s)#']
+
+Metrics/ModuleLength:
+  Max: 200
+
+Metrics/ClassLength:
+  Max: 200
+
+Metrics/MethodLength:
+  Max: 40
+
+Metrics/AbcSize:
+  Max: 60
+
+Metrics/BlockLength:
+  Exclude:
+    - 'spec/**/*'
+
+ParameterLists:
+  Max: 10
+
+Security/MarshalLoad:
+  Enabled: false
+
+Naming/UncommunicativeMethodParamName:
+  Enabled: false
+
+Style/FormatStringToken:
+  Enabled: false

data/.travis.yml

@@ -0,0 +1,11 @@
+sudo: false
+os: linux
+dist: trusty
+language: ruby
+rvm:
+  - 2.2
+  - 2.3
+  - 2.4
+  - 2.5
+before_install:
+  - gem install --no-document bundler -v '~> 1.16'

data/CODE_OF_CONDUCT.md

@@ -0,0 +1,74 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to making participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, gender identity and expression, level of experience,
+nationality, personal appearance, race, religion, or sexual identity and
+orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+  address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies both within project spaces and in public spaces
+when an individual is representing the project or its community. Examples of
+representing a project or community include using an official project e-mail
+address, posting via an official social media account, or acting as an appointed
+representative at an online or offline event. Representation of a project may be
+further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at yoshoku@outlook.com. All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at [http://contributor-covenant.org/version/1/4][version]
+
+[homepage]: http://contributor-covenant.org
+[version]: http://contributor-covenant.org/version/1/4/

data/Gemfile

@@ -0,0 +1,6 @@
+source 'https://rubygems.org'
+
+git_source(:github) { |repo_name| "https://github.com/#{repo_name}" }
+
+# Specify your gem's dependencies in hanny.gemspec
+gemspec

data/LICENSE.txt

@@ -0,0 +1,23 @@
+Copyright (c) 2017 yoshoku
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

data/README.md

@@ -0,0 +1,150 @@
+# Hanny
+
+[![Build Status](https://travis-ci.org/yoshoku/Hanny.svg?branch=master)](https://travis-ci.org/yoshoku/Hanny)
+[![Coverage Status](https://coveralls.io/repos/github/yoshoku/Hanny/badge.svg?branch=master)](https://coveralls.io/github/yoshoku/Hanny?branch=master)
+[![Gem Version](https://badge.fury.io/rb/hanny.svg)](https://badge.fury.io/rb/hanny)
+[![BSD 2-Clause License](https://img.shields.io/badge/License-BSD%202--Clause-orange.svg)](https://github.com/yoshoku/Hanny/blob/master/LICENSE.txt)
+
+Hanny is a Hash-based Approximate Nearest Neighbor (ANN) search library in Ruby.
+Hash-based ANN converts vector data into binary codes and builds a hash table by using the binary codes as hash keys.
+To build the hash table, Hanny uses Locality Sensitive Hashing (LSH) of approximating cosine similarity.
+It is known that if the code length is sufficiently long (ex. greater than 128-bit), LSH can obtain high search performance.
+In the experiment, Hanny achieved about twenty times faster search speed than the brute-force search by Euclidean distance.
+
+## Installation
+
+Add this line to your application's Gemfile:
+
+```ruby
+gem 'hanny'
+```
+
+And then execute:
+
+    $ bundle
+
+Or install it yourself as:
+
+    $ gem install hanny
+
+## Usage
+
+```ruby
+require 'hanny'
+
+# Prepare vector data for search targets and queries with Numo::DFloat (shape: [n_samples, n_features]).
+targets = Numo::DFloat.new(5000, 512).rand
+queries = Numo::DFloat.new(10, 512).rand
+
+# Build a search index with 256-bit binary code.
+index = Hanny::LSHIndex.new(code_length: 256)
+index.build_index(targets)
+
+# Obtain the Array<Integer> that has the data indices of 10-nearest neighbors for each query.
+candidates = index.search_knn(queries, n_neighbors: 10)
+
+# Obtain the Array<Integer> that has the data indices whithin Hamming radius of 4 for each query.
+candidates = index.search_radius(queries, radius: 4)
+
+# Calculate pairwise euclidean distances between the query and its neighbors.
+query_id = 0
+distances = Hanny::Utils.euclidean_distance(queries[query_id, true], targets[candidates[query_id], true])
+
+# Add new data to the search index.
+appended_data_ids = index.append_data(new_data)
+
+# Remove the data from the search index.
+removed_data_ids = index.remove_data([0, 1, 2])
+
+# Save and load the search index with Marshal.
+File.open('index.dat', 'wb') { |f| f.write(Marshal.dump(index)) }
+index = Marshal.load(File.binread('index.dat'))
+```
+
+## Experiment
+
+I confirmed the search speed of Hanny's LSH with [MNIST](https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass.html#mnist) data set.
+The experiment is carried out on MacBook Early 2016 (Core m3 1.1 GHz CPU and 8 GB memory).
+
+Code:
+```ruby
+require 'benchmark'
+require 'svmkit'
+require 'hanny'
+
+# Load MNIST data set.
+samples, labels = SVMKit::Dataset.load_libsvm_file('mnist')
+samples = Numo::DFloat.cast(samples)
+queries = samples[0..5, true]
+targets = samples[6..-1, true]
+qlabels = labels[0..5]
+tlabels = labels[6..-1]
+
+# Build LSH search index.
+index = Hanny::LSHIndex.new(code_length: 128, random_seed: 1)
+index.build_index(targets)
+
+# Run a benchmark test for finding 5-nearest neighbors.
+n_queries = queries.shape[0]
+n_neighbors = 5
+Benchmark.bm 50 do |r|
+  r.report 'LSH' do
+    candidates = index.search_knn(queries, n_neighbors: n_neighbors)
+    n_queries.times do |m|
+      STDERR.write("\nquery label: %d, neighbors label: " % qlabels[m])
+      candidates[m].each { |n| STDERR.write("%d, " % tlabels[n]) }
+    end
+    STDERR.write("\n")
+  end
+  r.report 'Brute-force' do
+    distance_mat = Hanny::Utils.euclidean_distance(queries, targets)
+    candidates = Array.new(n_queries) do |n|
+      distance_mat[n, true].to_a.map.with_index.sort_by(&:first).map(&:last)[0...n_neighbors]
+    end
+    n_queries.times do |m|
+      STDERR.write("\nquery label: %d, neighbors label: " % qlabels[m])
+      candidates[m].each { |n| STDERR.write("%d, " % tlabels[n]) }
+    end
+    STDERR.write("\n")
+  end
+end
+```
+
+Result:
+```bash
+    user     system      total        real
+LSH
+query label: 5, neighbors label: 5, 5, 5, 5, 5,
+query label: 0, neighbors label: 0, 0, 0, 0, 0,
+query label: 4, neighbors label: 4, 4, 4, 4, 4,
+query label: 1, neighbors label: 1, 1, 1, 1, 1,
+query label: 9, neighbors label: 9, 9, 9, 9, 9,
+query label: 2, neighbors label: 2, 2, 2, 2, 2,
+  0.290000   0.010000   0.300000 (  0.307445)
+Brute-force
+query label: 5, neighbors label: 5, 5, 5, 3, 5,
+query label: 0, neighbors label: 0, 0, 0, 0, 0,
+query label: 4, neighbors label: 4, 4, 4, 4, 4,
+query label: 1, neighbors label: 1, 1, 1, 1, 1,
+query label: 9, neighbors label: 9, 9, 9, 9, 9,
+query label: 2, neighbors label: 2, 2, 2, 2, 2,
+  6.350000   0.280000   6.630000 (  6.682365)
+```
+
+## Development
+
+After checking out the repo, run `bin/setup` to install dependencies. Then, run `rake spec` to run the tests. You can also run `bin/console` for an interactive prompt that will allow you to experiment.
+
+To install this gem onto your local machine, run `bundle exec rake install`. To release a new version, update the version number in `version.rb`, and then run `bundle exec rake release`, which will create a git tag for the version, push git commits and tags, and push the `.gem` file to [rubygems.org](https://rubygems.org).
+
+## Contributing
+
+Bug reports and pull requests are welcome on GitHub at https://github.com/yoshoku/Hanny. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the [Contributor Covenant](http://contributor-covenant.org) code of conduct.
+
+## License
+
+The gem is available as open source under the terms of the [BSD 2-clause License](https://opensource.org/licenses/BSD-2-Clause).
+
+## Code of Conduct
+
+Everyone interacting in the Hanny project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the [code of conduct](https://github.com/yoshoku/Hanny/blob/master/CODE_OF_CONDUCT.md).

data/Rakefile

@@ -0,0 +1,6 @@
+require 'bundler/gem_tasks'
+require 'rspec/core/rake_task'
+
+RSpec::Core::RakeTask.new(:spec)
+
+task default: :spec

data/bin/console

@@ -0,0 +1,14 @@
+#!/usr/bin/env ruby
+
+require 'bundler/setup'
+require 'hanny'
+
+# You can add fixtures and/or initialization code here to make experimenting
+# with your gem easier. You can also use a different console, if you like.
+
+# (If you use this, don't forget to add pry to your Gemfile!)
+# require "pry"
+# Pry.start
+
+require 'irb'
+IRB.start(__FILE__)

data/bin/setup

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+IFS=$'\n\t'
+set -vx
+
+bundle install
+
+# Do any other automated setup that you need to do here

data/hanny.gemspec

@@ -0,0 +1,38 @@
+
+lib = File.expand_path('lib', __dir__)
+$LOAD_PATH.unshift(lib) unless $LOAD_PATH.include?(lib)
+require 'hanny/version'
+
+Gem::Specification.new do |spec|
+  spec.name          = 'hanny'
+  spec.version       = Hanny::VERSION
+  spec.authors       = ['yoshoku']
+  spec.email         = ['yoshoku@outlook.com']
+
+  spec.summary       = 'Hanny is a Hash-based Approximate Nearest Neighbor search library in Ruby.'
+  spec.description   = <<MSG
+Hanny is a Hash-based Approximate Nearest Neighbor (ANN) search library in Ruby.
+Hash-based ANN converts vector data into binary codes and builds a hash table by using the binary codes as hash keys.
+To build the hash table, Hanny uses Locality Sensitive Hashing (LSH) of approximating cosine similarity.
+It is known that if the code length is sufficiently long (ex. greater than 128-bit), LSH can obtain high search performance.
+In the experiment, Hanny achieved about twenty times faster search speed than the brute-force search by Euclidean distance.
+MSG
+  spec.homepage      = 'https://github.com/yoshoku/hanny'
+  spec.license       = 'BSD-2-Clause'
+
+  spec.files         = `git ls-files -z`.split("\x0").reject do |f|
+    f.match(%r{^(test|spec|features)/})
+  end
+  spec.bindir        = 'exe'
+  spec.executables   = spec.files.grep(%r{^exe/}) { |f| File.basename(f) }
+  spec.require_paths = ['lib']
+
+  spec.required_ruby_version = '>= 2.1'
+
+  spec.add_runtime_dependency 'numo-narray', '>= 0.9.0'
+
+  spec.add_development_dependency 'bundler', '~> 1.16'
+  spec.add_development_dependency 'coveralls', '~> 0.8'
+  spec.add_development_dependency 'rake', '~> 10.0'
+  spec.add_development_dependency 'rspec', '~> 3.0'
+end

data/lib/hanny.rb

@@ -0,0 +1,8 @@
+# frozen_string_literal: true
+
+require 'zlib'
+require 'numo/narray'
+
+require 'hanny/version'
+require 'hanny/utils'
+require 'hanny/lsh_index'

data/lib/hanny/lsh_index.rb

@@ -0,0 +1,273 @@
+# frozen_string_literal: true
+
+module Hanny
+  # LSHIndex is a class that builds a search index with Locality Sensitive Hashing (LSH) [1].
+  # It is known that if the code length is sufficiently long (ex. greater than 128-bit),
+  # LSH can obtain higher search performance than many popular hashing methods [2].
+  # In search process, LSHIndex obtains search results by sorting the data stored in hash table with Hamming distances
+  # between query binary code and binary hash keys.
+  #
+  # @example
+  #   # Prepare vector data for search targets and queries with Numo::DFloat (shape: [n_samples, n_features]).
+  #   targets = Numo::DFloat.new(5000, 512).rand
+  #   queries = Numo::DFloat.new(10, 512).rand
+  #
+  #   # Build a search index with 256-bit binary code via LSH.
+  #   # Although LSHIndex works without setting random_seed, it recommends setting random_seed for reproducibility.
+  #   index = Hanny::LSHIndex.new(code_length: 256, random_seed: 1)
+  #   index.build_index(targets)
+  #
+  #   # Obtain the Array<Integer> that has the data indices of 10-neighbors for each query.
+  #   candidates = index.search_knn(queries, n_neighbors: 10)
+  #
+  #   # Save and load the search index with Marshal.
+  #   File.open('index.dat', 'wb') { |f| f.write(Marshal.dump(index)) }
+  #   index = Marshal.load(File.binread('index.dat'))
+  #
+  # *References:*
+  # 1. Moses S. Charikar, "Similarity Estimation Techniques from Rounding Algorithms," Proc. of the 34-th Annual ACM Symposium on Theory of Computing, pp. 380--388, (2002).
+  # 1. Deng Cai, "A Revisit of Hashing Algorithms for Approximate Nearest Neighbor Search," CoRR abs/1612.07545 (2016).
+  class LSHIndex
+    # Return the code length of hash key.
+    # @return [Integer]
+    attr_reader :code_length
+
+    # Return the number of samples of indexed data.
+    # @return [Integer]
+    attr_reader :n_samples
+
+    # Return the number of features of indexed data.
+    # @return [Integer]
+    attr_reader :n_features
+
+    # Return the number of hash keys.
+    # @return [Integer]
+    attr_reader :n_keys
+
+    # Return the hash table.
+    # @return [Hash]
+    attr_reader :hash_table
+
+    # Return the binary hash codes.
+    # @return [Numo::Bit]
+    attr_reader :hash_codes
+
+    # Return the seed to initialize random number generator.
+    # @return [Integer]
+    attr_reader :random_seed
+
+    # Return the random generator to generate random matrix.
+    # @return [Random]
+    attr_reader :rng
+
+    # Create a new nearest neighbor index.
+    # @param code_length [Integer] The length of binary code for hash key.
+    # @param random_seed [Integer/NilClass] The seed value using to initialize the random generator.
+    def initialize(code_length: 256, random_seed: nil)
+      @code_length = code_length
+      @n_samples = nil
+      @n_features = nil
+      @n_keys = nil
+      @last_id = nil
+      @weight_mat = nil
+      @hash_table = nil
+      @hash_codes = nil
+      @random_seed = random_seed
+      @random_seed ||= srand
+      @rng = Random.new(@random_seed)
+    end
+
+    # Convert data into binary codes.
+    # @param x [Numo::DFloat] (shape: [n_samples, n_features]) The data to be converted to binary codes.
+    # @return [Numo::Bit] The binary codes converted from given data.
+    def hash_function(x)
+      x.dot(@weight_mat).ge(0.0)
+    end
+
+    # Build a search index.
+    # @param x [Numo::DFloat] (shape: [n_samples, n_features]) The dataset for building search index.
+    # @return [SVC] The search index itself that has constructed the hash table.
+    def build_index(x)
+      # Initialize some variables.
+      @n_samples, @n_features = x.shape
+      @hash_table = {}
+      @hash_codes = []
+      @weight_mat = Utils.rand_normal([@n_features, @code_length], @rng)
+      # Convert samples to binary codes.
+      bin_x = hash_function(x)
+      # Store samples to binary hash table.
+      @n_samples.times do |m|
+        bin_code = bin_x[m, true]
+        hash_key = symbolized_hash_key(bin_code)
+        unless @hash_table.key?(hash_key)
+          @hash_codes.push(bin_code.to_a)
+          @hash_table[hash_key] = []
+        end
+        @hash_table[hash_key].push(m)
+      end
+      @hash_codes = Numo::Bit.cast(@hash_codes)
+      # Update some variables.
+      @n_keys = @hash_codes.shape[0]
+      @last_id = @n_samples
+      self
+    end
+
+    # Append new data to the search index.
+    # @param x [Numo::DFloat] (shape: [n_samples, n_features]) The dataset to append to search index.
+    # @return [Array<Integer>] The indices of appended data in search index
+    def append_data(x)
+      # Initialize some variables.
+      n_new_samples, = x.shape
+      bin_x = hash_function(x)
+      added_data_ids = []
+      # Store samples to binary hash table.
+      new_codes = []
+      n_new_samples.times do |m|
+        bin_code = bin_x[m, true]
+        hash_key = symbolized_hash_key(bin_code)
+        unless @hash_table.key?(hash_key)
+          new_codes.push(bin_code.to_a)
+          @hash_table[hash_key] = []
+        end
+        new_data_id = @last_id + m
+        @hash_table[hash_key].push(new_data_id)
+        added_data_ids.push(new_data_id)
+      end
+      # Update hash codes.
+      unless new_codes.empty?
+        new_codes = Numo::Bit.cast(new_codes)
+        @hash_codes = @hash_codes.concatenate(new_codes)
+        @n_keys = @hash_codes.shape[0]
+      end
+      @last_id += n_new_samples
+      @n_samples += n_new_samples
+      added_data_ids
+    end
+
+    # Remove data from the search index.
+    # The indices of removed data will never be assigned unless the search index is rebuilt.
+    # @param data_ids [Array<Integer>] The data indices to be removed.
+    # @return [Array<Integer>] The indices of removed data in search index
+    def remove_data(data_ids)
+      removed_data_ids = []
+      data_ids.each do |query_id|
+        # Remove data id from hash table.
+        hash_key = @hash_table.keys.select { |k| @hash_table[k].include?(query_id) }.first
+        next if hash_key.nil?
+        @hash_table[hash_key].delete(query_id)
+        removed_data_ids.push(query_id)
+        # Remove the hash key if there is no data.
+        next unless @hash_table[hash_key].empty?
+        target_id = distances_to_hash_codes(decoded_hash_key(hash_key)).index(0)
+        @hash_codes = @hash_codes.delete(target_id, 0)
+      end
+      @n_samples -= removed_data_ids.size
+      removed_data_ids
+    end
+
+    # Perform k-nearest neighbor search.
+    # @param q [Numo::DFloat] (shape: [n_queries, n_features]) The data for search queries.
+    # @param n_neighbors [Integer] The number of neighbors.
+    # @return [Array<Integer>] The data indices of search result.
+    def search_knn(q, n_neighbors: 10)
+      # Initialize some variables.
+      n_queries, = q.shape
+      candidates = Array.new(n_queries) { [] }
+      # Binarize queries.
+      bin_q = hash_function(q)
+      # Find k-nearest neighbors for each query.
+      n_queries.times do |m|
+        sort_with_index(distances_to_hash_codes(bin_q[m, true])).each do |_, n|
+          candidates[m] = candidates[m] | @hash_table[symbolized_hash_key(@hash_codes[n, true])]
+          break if candidates[m].size >= n_neighbors
+        end
+        candidates[m] = candidates[m].shift(n_neighbors)
+      end
+      candidates
+    end
+
+    # Perform hamming radius nearest neighbor search.
+    # @param q [Numo::DFloat] (shape: [n_queries, n_features]) The data for search queries.
+    # @param radius [Float] The hamming radius for search range.
+    # @return [Array<Integer>] The data indices of search result.
+    def search_radius(q, radius: 1)
+      # Initialize some variables.
+      n_queries, = q.shape
+      candidates = Array.new(n_queries) { [] }
+      # Binarize queries.
+      bin_q = hash_function(q)
+      # Find k-nearest neighbors for each query.
+      n_queries.times do |m|
+        sort_with_index(distances_to_hash_codes(bin_q[m, true])).each do |d, n|
+          break if d > radius
+          candidates[m] = candidates[m] | @hash_table[symbolized_hash_key(@hash_codes[n, true])]
+        end
+      end
+      candidates
+    end
+
+    # Dump marshal data.
+    # @return [Hash] The marshal data for search index.
+    def marshal_dump
+      { code_length: @code_length,
+        n_samples: @n_samples,
+        n_features: @n_features,
+        n_keys: @n_keys,
+        last_id: @last_id,
+        weight_mat: @weight_mat,
+        bias_vec: @bias_vec,
+        hash_table: @hash_table,
+        hash_codes: @hash_codes,
+        random_seed: @random_seed,
+        rng: @rng }
+    end
+
+    # Load marshal data.
+    # @return [nil]
+    def marshal_load(obj)
+      @code_length = obj[:code_length]
+      @n_samples = obj[:n_samples]
+      @n_features = obj[:n_features]
+      @n_keys = obj[:n_keys]
+      @last_id = obj[:last_id]
+      @weight_mat = obj[:weight_mat]
+      @bias_vec = obj[:bias_vec]
+      @hash_table = obj[:hash_table]
+      @hash_codes = obj[:hash_codes]
+      @random_seed = obj[:random_seed]
+      @rng = obj[:rng]
+      nil
+    end
+
+    private
+
+    # Convert binary code to symbol as hash key.
+    # @param bin_code [Numo::Bit]
+    # @return [Symbol]
+    def symbolized_hash_key(bin_code)
+      Zlib::Deflate.deflate(bin_code.to_a.join, Zlib::BEST_SPEED).to_sym
+    end
+
+    # Calculate hamming distances between binary code and binary hash keys.
+    # @param bin_code [Numo::Bit]
+    # @return [Array<Float>]
+    def distances_to_hash_codes(bin_code)
+      (bin_code ^ @hash_codes).count(1).to_a
+    end
+
+    # Sort array elements with indices.
+    # @param arr [Array<Float>]
+    # @return [Array<Float, Integer>]
+    def sort_with_index(arr)
+      arr.map.with_index.sort_by(&:first)
+    end
+
+    # Convert hash key symbol to binary code.
+    # @param hash_key [Symbol]
+    # @return [Numo::Bit]
+    def decoded_hash_key(hash_key)
+      bin_code = Zlib::Inflate.inflate(hash_key.to_s).split('').map(&:to_i)
+      Numo::Bit[*bin_code]
+    end
+  end
+end

data/lib/hanny/utils.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+module Hanny
+  # This module consists of utility methods.
+  module Utils
+    class << self
+      # Calculate pairwise euclidean distances between x and y.
+      # @param x [Numo::DFloat] (shape: [n_samples_x, n_features])
+      # @param y [Numo::DFloat] (shape: [n_samples_y, n_features])
+      # @return [Numo::DFloat] (shape: [n_samples_x, n_samples_x] or [n_samples_x, n_samples_y] if y is given)
+      def euclidean_distance(x, y = nil)
+        y = x if y.nil?
+        x = Numo::DFloat[x] if x.shape[1].nil?
+        y = Numo::DFloat[y] if y.shape[1].nil?
+        sum_x_vec = (x**2).sum(1)
+        sum_y_vec = (y**2).sum(1)
+        dot_xy_mat = x.dot(y.transpose)
+        distance_matrix = dot_xy_mat * -2.0 +
+                          sum_x_vec.tile(y.shape[0], 1).transpose +
+                          sum_y_vec.tile(x.shape[0], 1)
+        Numo::NMath.sqrt(distance_matrix.abs)
+      end
+
+      # Generate a uniform random matrix with random number generator.
+      # @param shape [Array<Integer>] The size of random matrix.
+      # @param rng [Random] The random number generator
+      # @return [Numo::DFloat] The generated uniform random matrix.
+      def rand_uniform(shape, rng)
+        rnd_vals = Array.new(shape.inject(:*)) { rng.rand }
+        Numo::DFloat.asarray(rnd_vals).reshape(shape[0], shape[1])
+      end
+
+      # Generate a normal random matrix with random number generator.
+      # @param shape [Array<Integer>] The size of random matrix.
+      # @param rng [Random] The random number generator
+      # @return [Numo::DFloat] The generated normal random matrix.
+      def rand_normal(shape, rng, mu = 0.0, sigma = 1.0)
+        a = rand_uniform(shape, rng)
+        b = rand_uniform(shape, rng)
+        (Numo::NMath.sqrt(Numo::NMath.log(a) * -2.0) * Numo::NMath.sin(b * 2.0 * Math::PI)) * sigma + mu
+      end
+    end
+  end
+end

data/lib/hanny/version.rb

@@ -0,0 +1,7 @@
+# frozen_string_literal: true
+
+# Hanny is a hash-based approximate nearest neighbor search library.
+module Hanny
+  # @!visibility private
+  VERSION = '0.1.0'.freeze
+end

metadata

@@ -0,0 +1,136 @@
+--- !ruby/object:Gem::Specification
+name: hanny
+version: !ruby/object:Gem::Version
+  version: 0.1.0
+platform: ruby
+authors:
+- yoshoku
+autorequire: 
+bindir: exe
+cert_chain: []
+date: 2018-05-04 00:00:00.000000000 Z
+dependencies:
+- !ruby/object:Gem::Dependency
+  name: numo-narray
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: 0.9.0
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: 0.9.0
+- !ruby/object:Gem::Dependency
+  name: bundler
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.16'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.16'
+- !ruby/object:Gem::Dependency
+  name: coveralls
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '0.8'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '0.8'
+- !ruby/object:Gem::Dependency
+  name: rake
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '10.0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '10.0'
+- !ruby/object:Gem::Dependency
+  name: rspec
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '3.0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '3.0'
+description: |
+  Hanny is a Hash-based Approximate Nearest Neighbor (ANN) search library in Ruby.
+  Hash-based ANN converts vector data into binary codes and builds a hash table by using the binary codes as hash keys.
+  To build the hash table, Hanny uses Locality Sensitive Hashing (LSH) of approximating cosine similarity.
+  It is known that if the code length is sufficiently long (ex. greater than 128-bit), LSH can obtain high search performance.
+  In the experiment, Hanny achieved about twenty times faster search speed than the brute-force search by Euclidean distance.
+email:
+- yoshoku@outlook.com
+executables: []
+extensions: []
+extra_rdoc_files: []
+files:
+- ".coveralls.yml"
+- ".gitignore"
+- ".rspec"
+- ".rubocop.yml"
+- ".travis.yml"
+- CODE_OF_CONDUCT.md
+- Gemfile
+- LICENSE.txt
+- README.md
+- Rakefile
+- bin/console
+- bin/setup
+- hanny.gemspec
+- lib/hanny.rb
+- lib/hanny/lsh_index.rb
+- lib/hanny/utils.rb
+- lib/hanny/version.rb
+homepage: https://github.com/yoshoku/hanny
+licenses:
+- BSD-2-Clause
+metadata: {}
+post_install_message: 
+rdoc_options: []
+require_paths:
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '2.1'
+required_rubygems_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+requirements: []
+rubyforge_project: 
+rubygems_version: 2.7.6
+signing_key: 
+specification_version: 4
+summary: Hanny is a Hash-based Approximate Nearest Neighbor search library in Ruby.
+test_files: []