Skip to main content

Fast and customizable tokenizer

Project description

## tok

[![PyPI](]( [![PyPI](](

Fast and most complete/customizable tokenizer in Python.

It is roughly 25x faster than spacy’s and nltk’s regex based tokenizers.

Using the aho-corasick algorithm makes it a novelty and allows it to be both explainable and fast in how it will split.

The heavy lifting is done by [textsearch]( and [pyahocorasick](, allowing this to be written in only ~200 lines of code.

Contrary to regex-based approaches, it will go over each character in a text only once. Read [below](#how-it-works) about how this works.

### Installation

pip install tok

### Usage

By default it handles contractions, http, (float) numbers and currencies.

`python from tok import word_tokenize word_tokenize("I wouldn't do that.... would you?") ['I', 'would', 'not', 'do', 'that', '...', 'would', 'you', '?'] `

Or configure it yourself:

`python from tok import Tokenizer tokenizer = Tokenizer(protected_words=["some.thing"]) # still using the defaults tokenizer.word_tokenize("I want to protect some.thing") ['I', 'want', 'to', 'protect', 'some.thing'] `

Split by sentences:

`python from tok import sent_tokenize sent_tokenize("I wouldn't do that.... would you?") [['I', 'would', 'not', 'do', 'that', '...'], ['would', 'you', '?']] `

for more options check the documentation of the Tokenizer.

### Further customization


`python from tok import Tokenizer t = Tokenizer(protected_words=["some.thing"]) # still using the defaults `

You can add your own ideas to the tokenizer by using:

  • t.keep(x, reason): Whenever it finds x, it will not add whitespace. Prevents direct tokenization.

  • t.split(x, reason): Whenever it finds x, it will surround it by whitespace, thus creating a token.

  • t.drop(x, reason): Whenever it finds x, it will remove it but add a split.

  • t.strip(x, reason): Whenever it finds x, it will remove it without splitting.

`python t.drop("bla", "bla is not needed") t.word_tokenize("Please remove bla, thank you") ['Please', 'remove', ',', 'thank', 'you'] `

### Explainable

Explain what happened:

`python t.explain("bla") [{'from': 'bla', 'to': ' ', 'explanation': 'bla is not needed'}] `

See everything in there (will help you understand how it works):

`python t.explain_dict `

### How it works

It will always only keep the longest match. By introducing a space in your tokens, it will make it be split.

If you consider how the tokenization of . works, see here:

  • When it finds a ` A.` it will make it ` A.` (single letter abbreviations)

  • When it finds a .0 it will make it .0 (numbers)

  • When it finds a ., it will make it ` . ` (thus making a split)

If you want to make sure something including a dot stays, you can use for example:


### Contributing

It would be greatly appreciated if you want to contribute to this library.

It would also be great to add [contractions]( for other languages.

Download files

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

Source Distribution

tok-0.1.14.tar.gz (5.6 kB view hashes)

Uploaded source

Built Distribution

tok-0.1.14-py2.py3-none-any.whl (7.8 kB view hashes)

Uploaded 3 7

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page