SubwordTextEncoder Library
C++ implemented Subword Text Encoder for Natural Language Processing.
Motivations & Inspirations
- SubwordTextEncoder
- Originally Created for speeding up the Gavin Project.
- I wanted to properly test out C++, figured re-creating something like this is the way to go.
So how does it work?
Well, the reason I use a Subword Text Encoder(STE), over a regular Encoder is because STEs are able to encode words they've never seen before. Through Byte encoding. If the STE is given a word not in its vocabulary then it splits the word into letters and encodes each individual letter. Therefore while models using the STE cannot come up with these words alone, they can understand them and link them to other ones.
How to use this
The STE, is a simple class inside the namespace "tokenizers" (this is so I can add more tokenizers in the future if I feel the need.), The actual STE class is called "SubwordTextEncoder".
Constructors
To create an instance there are 2 ways.
tokenizers::SubwordTextEncoder(target_vocab_size, name)ortokenizer::SubwordTextEncoder(filename_prefix)
For the first Constructor, you pass 2 variables, first one decides how large the vocabulary size should be, while the name is just a way to keep track of different STEs. The second method, allows you to load a previously saved STE.
Generating a vocabulary
Assuming you haven't loaded a previous STE, you can choose to generate a vocabulary
through its method SubwordTextEncoder.build_vocabulary(texts) this methods
takes a list of strings, each one a sentence. From this it gets the most common,
words and add those to the vocabulary. This is the biggest difference from TF's
implementation and also the major draw back. Note: This is a mutli-threaded method,
it will try to use as much of your CPU as it can.
Encoding & Decoding
Whether you have loaded, or built your own vocabulary the process for encoding & decoding
is the same. To encode a string just pass it to the encode method
SubwordTextEncoder.encode(string), this will return a list of integer ID's.
To decode a list of ID's merely pass it into the decode method.
SubwordTextEncoder.decode(string) NOTE: All ID's are +1 its value inside the vocabulary
to allow for the "0" ID to be used as padding inside a model.
Licence
Tokenizer Library:
The class is licensed under the MIT Licence:
Copyright © 2021 Joshua Shiells
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
JSON Library used:
The class is licensed under the MIT License:
Copyright © 2013-2021 Niels Lohmann
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.