The fastent Python library is a tool for end-to-end creation of custom models for named-entity recognition.

To train a model for a new type of entity, you just need a list of examples.

You are not limited to only predefined types like person, location and organization.

fastent does end-to-end creation: dataset generation, annotation, contextualiziation and training a model.

You can also use fastent modules as standalone tools.

fastent includes integrations with tools like spaCy, fastText pre-trained models and NLTK.

fastent is built to scale to very large text datasets in many languages.

• Installation
• How To
  • Dataset Generation
  • Annotation
  • Contextualization
  • Training
  • Testing
• Integrations
  • Pre-trained Models
  • Text utilities
  • WordNet
  • Poincaré embeddings
• More

fastent is developed for Python 3 on Unix systems.

Clone this repo or install from PyPI:

pip install fastent

Download NLTK data:

python -m nltk.downloader stopwords

Install and set up CouchDB:

wget -O - https://raw.githubusercontent.com/fastent/fastent/master/install.sh | bash

TODO: fastText stuff

fastent can generate a dataset from a list

TODO

fastent can even generate a list from one or two examples.

from fastent import dataset_pseudo_generator

model = dataset_pseudo_generator.spacy_initialize('en_core_web_lg')
dataset_pseudo_generator.dataset_generate(model,['cocaine', 'heroin'], 100)

The equivalent on the command line:

python dataset_pseudo_generator.py -m en_core_web_lg -s cocaine,heroin

TODO

TODO

To train a model from the annotated and contextualized dataset:

For now the only supported learning framework is spaCy.

Request support for a new learning framework

TODO: sample output

Coming soon!

fastent includes integrations for downloading datasets and pre-trained models.

TODO

See how fastent performs on benchmarks

Try the tutorial or fork examples

Browse frequently asked questions

Report bugs or request new features