A One Hot Vector: Difference between revisions
From Algolit
| (8 intermediate revisions by the same user not shown) | |||
| Line 8: | Line 8: | ||
|} | |} | ||
| − | + | A '''one-hot-vector''' is a word-representation technique that uses ''distributional similarity'' to find patterns in the phrases that are used in company of a word. One-hot-vectors are basically a big matrix of 0's, with as many rows and columns as there are unique words. A text with 500 unique words, will be represented by a matrix of 500x500. With this matrix as its central tool, a script will go through the sentences of a dataset and count how often a word appears next to another word. | |
| + | ==Recipe for a one hot vector== | ||
If this is our example sentence ... | If this is our example sentence ... | ||
| − | |||
"The algoliterary explorers discovered a multidimensional landscape made of words disguised as numbers." | "The algoliterary explorers discovered a multidimensional landscape made of words disguised as numbers." | ||
| − | |||
... these are the 14 words we work with ... | ... these are the 14 words we work with ... | ||
| − | |||
a | a | ||
algoliterary | algoliterary | ||
| Line 33: | Line 31: | ||
words | words | ||
. | . | ||
| − | |||
... a single vector in a one-hot-vector looks like this ... | ... 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] | [0 0 0 0 0 0 0 0 0 0 0 0 0 0] | ||
| − | |||
... and a full fourteen-dimensional matrix like this ... | ... 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] 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] algoliterary | ||
| Line 58: | Line 52: | ||
[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] words | ||
[0 0 0 0 0 0 0 0 0 0 0 0 0 0]] . | [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. | ... with one 0 for each unique word in a vocabulary, and a row for each unique word. | ||
| Line 64: | Line 57: | ||
The following step is to count how often a word appears next to another ... | 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." | "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 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 0 1 0 0 0 0 0 1 0 0] algoliterary | ||
| Line 83: | Line 73: | ||
[0 0 0 0 1 0 0 0 0 0 1 0 0 0] words | [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]] . | [0 0 0 0 0 0 0 0 0 1 0 0 0 0]] . | ||
| − | |||
| − | |||
| − | |||
| + | ==Algolit's 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: [https://gitlab.constantvzw.org/algolit/algolit/blob/master/algoliterary_encounter/one-hot-vector/one-hot-vector_gijs.py one-hot-vector_gijs.py] & [https://gitlab.constantvzw.org/algolit/algolit/blob/master/algoliterary_encounter/one-hot-vector/one-hot-vector_hans.py one-hot-vector_hans.py] | 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: [https://gitlab.constantvzw.org/algolit/algolit/blob/master/algoliterary_encounter/one-hot-vector/one-hot-vector_gijs.py one-hot-vector_gijs.py] & [https://gitlab.constantvzw.org/algolit/algolit/blob/master/algoliterary_encounter/one-hot-vector/one-hot-vector_hans.py one-hot-vector_hans.py] | ||
| − | =Note that= | + | ==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, | + | "''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, [https://cs224d.stanford.edu/lectures/CS224d-Lecture2.pdf#6 CS224d, Deep Learning for Natural Language Processing at Stanford University (2016)].) |
For more notes on this lecture visit http://pad.constantvzw.org/public_pad/neural_networks_3 | For more notes on this lecture visit http://pad.constantvzw.org/public_pad/neural_networks_3 | ||
| − | |||
| − | |||
[[Category:Algoliterary-Encounters]] | [[Category:Algoliterary-Encounters]] | ||
Latest revision as of 22:05, 31 October 2017
| Type: | Algoliterary exploration |
| Technique: | word-embeddings |
| Developed by: | Algolit |
A one-hot-vector is a word-representation technique that uses distributional similarity to find patterns in the phrases that are used in company of a word. One-hot-vectors are basically a big matrix of 0's, with as many rows and columns as there are unique words. A text with 500 unique words, will be represented by a matrix of 500x500. With this matrix as its central tool, a script will go through the sentences of a dataset and count how often a word appears next to another word.
Recipe for a one hot vector
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's 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, CS224d, Deep Learning for Natural Language Processing at Stanford University (2016).)
For more notes on this lecture visit http://pad.constantvzw.org/public_pad/neural_networks_3