Ruby GloVe

Build Status Code Climate Test Coverage Inline docs

Ruby implementation of Global Vectors for Word Representations.


GloVe is an unsupervised learning algorithm for obtaining vector representations for words. Training is performed on aggregated global word-word co-occurrence statistics from a corpus, and the resulting representations showcase interesting linear substructures of the word vector space.

NOTE This is an early prototype.



This library relies on the rb-gsl gem for Matrix and Vector operations, therefore you need GNU Scientific Library installed.


$ sudo apt-get install libgsl0-dev


$ brew install gsl

Only compatible with MRI: tested in versions 2.0.x and 2.1.x


$ gem install glove

or add to your Gemfile

gem 'glove'


require 'glove'

# See documentation for all available options
model =

# Next feed it some text.
text ='quantum-physics.txt')

# Or you can pass it a Glove::Corpus object as the text argument instead
corpus =

# Finally, to query the model, we need to train it

# So far, word similarity and analogy task methods have been included:
# Most similar words to quantum
# => [["physic", 0.9974459436353388], ["mechan", 0.9971606266531394], ["theori", 0.9965966776283189]]

# What words relate to atom like quantum relates to physics?
model.analogy_words('quantum', 'physics', 'atom')
# => [["electron", 0.9858380292886947], ["energi", 0.9815122410243475], ["photon", 0.9665073849076669]]

# Save the trained matrices and vectors for later usage in binary formats'corpus.bin', 'cooc-matrix.bin', 'word-vec.bin', 'word-biases.bin')

# Later on create a new instance and call #load
model =
model.load('corpus.bin', 'cooc-matrix.bin', 'word-vec.bin', 'word-biases.bin')
# Now you can query the model again and get the same results as above


Thanks to the rb-gsl bindings for GSL, matrix/vector operations are fast. The glove algorythm itself, however, requires quite a bit of computational power, even the original C library. If you need speed, use smaller texts with vocabulaty size no more than 100K words. Processing text with 160K words (compilation of several books on quantum mechanics) on a late 2012 MBP (8GB RAM) with ruby-2.1.5 takes about 7 minutes:

 $ ruby -Ilib benchmark/benchmark.rb
                 user     system      total        real
Fit Text    11.320000   0.070000  11.390000 ( 11.387612)
Vocabulary size: 158323
Unique tokens: 2903
Co-occur     1.330000   0.250000 1107.720000 (300.738453)
Train      121.120000  12.960000  134.080000 (128.409034)
Similarity   0.010000   0.000000    0.010000 (  0.057423)
Give me the 3 most similar words to quantum
[["problem", 0.9977609386134489], ["mechan", 0.9977529272587808], ["classic", 0.9974759411408415]]
Analogy      0.010000   0.000000   0.010000 (  0.010674)
What 3 words relate to atom like quantum relates to mechanics?
[["particl", 0.9982711579369483], ["find", 0.9982303885530384], ["expect", 0.9982017117355527]]


  • Word Vector graphs


  1. Fork it ( )
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Add some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create a new Pull Request