A One Hot Vector

Type:	Algoliterary exploration
Technique:	word-embeddings
Developed by:	Algolit

"Meaning is this illusive thing that were trying to capture" (Richard Socher in CS224D Lecture 2 - 31st Mar 2016 (Youtube))

Word embeddings are used to represent words as inputs to machine learning. The words become vectors in a multi-dimensional space, where nearby vectors represent similar meanings. With word embeddings, you can compare words by (roughly) what they mean, not just exact string matches.

Successfully training word vectors requires starting from hundreds of gigabytes of input text. Fortunately, various machine-learning groups have already done this and provided pre-trained word embeddings that one can download. Two very well-known datasets of pre-trained English word embeddings are word2vec, pre-trained on Google News data, and GloVe, pre-trained on the Common Crawl of web pages.

The term has only recently entered the vocabulary of machine learning, with the expansion of the deep learning community. In computational linguistics the expression 'distributional semantic model' is sometimes preferred. Other terms include 'distributed representation', 'semantic vector space', or 'word space'.

Two popular examples of standalone implementations are the word2vec library (a single layered neural network) and the GloVe library (distributional semantic model).

Making a one-hot-vectors

If this is our example sentence ...

"The algoliterary explorers discovered a multidimensional landscape made of words disguised as numbers."

... these are the 14 words we work with ...

a
algoliterary
as
discovered
disguised
explores
landscape
made
multidimensional
numbers
of
the
words
.

... a single vector in a one-hot-vector looks like this ...

[0 0 0 0 0 0 0 0 0 0 0 0 0 0] 

... and a full fourteen-dimensional matrix like this ...

[[0 0 0 0 0 0 0 0 0 0 0 0 0 0]  a
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  algoliterary
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  as
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  discovered
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  disguised
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  explores
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  landscape
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  made
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  multidimensional
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  numbers
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  of
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  the
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]  words
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0]] .

... with one 0 for each unique word in a vocabulary, and a row for each unique word.

The following step is to count how often a word appears next to another ...

"The algoliterary explorers discovered a multidimensional landscape made of words disguised as numbers."

[[0 0 0 1 0 0 0 0 1 0 0 0 0 0]  a
 [0 0 0 0 0 1 0 0 0 0 0 1 0 0]  algoliterary
 [0 0 0 0 1 0 0 0 0 1 0 0 0 0]  as
 [1 0 0 0 0 1 0 0 0 0 0 0 0 0]  discovered
 [0 0 1 0 0 0 0 0 0 0 0 0 1 0]  disguised
 [0 1 0 1 0 0 0 0 0 0 0 0 0 0]  explores
 [0 0 0 0 0 0 0 1 1 0 0 0 0 0]  landscape
 [0 0 0 0 0 0 1 0 0 0 1 0 0 0]  made
 [1 0 0 0 0 0 1 0 0 0 0 0 0 0]  multidimensional
 [0 0 1 0 0 0 0 0 0 0 0 0 0 1]  numbers
 [0 0 0 0 0 0 0 1 0 0 0 0 1 0]  of
 [0 1 0 0 0 0 0 0 0 0 0 0 0 0]  the
 [0 0 0 0 1 0 0 0 0 0 1 0 0 0]  words
 [0 0 0 0 0 0 0 0 0 1 0 0 0 0]] .

Algolit one-hot-vector scripts

Two one-hot-vector scripts were created during one of the Algolit sessions, both creating the same matrix but in a different way. To download and run them, use the following links: one-hot-vector_gijs.py & one-hot-vector_hans.py

Note that

"Words are represented once in a vector. So words with multiple meanings, like "bank", are more difficult to represent. There is research to multivectors for one word, so that it does not end up in the middle." (Richard Socher, idem.)]

For more notes on this lecture visit http://pad.constantvzw.org/public_pad/neural_networks_3