Skip to main content

A library that converts words to vectors using PMI and SVD

Project description


SVD2vec is a python library for representing documents words as vectors. Vectors are created using the PMI (Pointwise Mutual Information) and the SVD (Singular Value Decomposition).

This library implements recommendations from "Improving Distributional Similarity with Lessons Learned from Word Embeddings" (Omer Levy, Yoav Goldberg, and Ido Dagan). This papers suggests that traditional methods like PMI and SVD can be as good as word2vec by appling the same hyperparameters.

Documentation can be found at


pip install svd2vec


wget -O text8.gz
gzip -d text8.gz -f
# Building
>>> from svd2vec import svd2vec
>>> documents = [open("text8", "r").read().split(" ")]
>>> svd = svd2vec(documents, window=2, min_count=100)
# I/O
>>> svd = svd2vec.load("svd.bin")
# Similarities
>>> svd.similarity("bad", "good")
# 0.4156516999158368
>>> svd.similarity("monday", "friday")
# 0.839529117681973
# Most similar words
>>> svd.most_similar(positive=["january"], topn=2)
# [('february', 0.6854849518368631), ('october', 0.6653385092683669)]
>>> svd.most_similar(positive=['moscow', 'france'], negative=['paris'], topn=4)
# [('russia', 0.6221746629754187), ('ussr', 0.6024809889985986), ('soviet', 0.5794180517326273), ('bolsheviks', 0.5365123080505297)]
# Analogies
>>> svd.analogy("paris", "france", "berlin")
# [('germany', 0.6977716641680641), ...]
>>> svd.analogy("road", "cars", "rail")
# [('trains', 0.7532519174901262), ...]
>>> svd.analogy("cow", "cows", "pig")
# [('pigs', 0.6944101149919422), ...]
>>> svd.analogy("man", "men", "woman")
# [('women', 0.7471792753875327), ...]

Using Gensim you can load a svd2vec model using it's word2vec representation:

>>> from gensim.models.keyedvectors import Word2VecKeyedVectors
>>> svd.save_word2vec_format("svd_word2vec_format.txt")
>>> keyed_vector = Word2VecKeyedVectors.load_word2vec_format("svd_word2vec_format.txt")
>>> keyed_vector.similarity("good", "bad")
# 0.54922897

Improving Distributional Similarity with Lessons Learned from Word Embeddings
Omer Levy, Yoav Goldberg, and Ido Dagan
Transactions of the Association for Computational Linguistics 2015 Vol. 3, 211-225

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for svd2vec, version 0.3.3
Filename, size File type Python version Upload date Hashes
Filename, size svd2vec-0.3.3-py3-none-any.whl (174.5 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size svd2vec-0.3.3.tar.gz (173.7 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page