LightNER

Check Our New NER Toolkit🚀🚀🚀

• Inference:
  • LightNER: inference w. models pre-trained / trained w. any following tools, efficiently.
• Training:
  • LD-Net: train NER models w. efficient contextualized representations.
  • VanillaNER: train vanilla NER models w. pre-trained embedding.
• Distant Training:
  • AutoNER: train NER models w.o. line-by-line annotations and get competitive performance.

---

This package supports to conduct inference with models pre-trained by:

• Vanilla_NER: vanilla sequence labeling models.
• LD-Net: sequence labeling models w. efficient contextualized representation.
• AutoNER: distant supervised named entity recognition models (no line-by-line annotations for training).

We are in an early-release beta. Expect some adventures and rough edges.

Quick Links

• Installation
• Usage

Installation

To install via pypi:

pip install lightner

To build from source:

pip install git+https://github.com/LiyuanLucasLiu/LightNER

or

git clone https://github.com/LiyuanLucasLiu/LightNER.git
cd LightNER
python setup.py install

Usage

Pre-trained Models

	Model	Task	Performance
LD-Net	pner1.th	NER for (PER, LOC, ORG & MISC)	F1 92.21
LD-Net	pnp0.th	Chunking	F1 95.79
Vanilla_NER		NER for (PER, LOC, ORG & MISC)
Vanilla_NER		Chunking
AutoNER	autoner0.th	Distant NER trained w.o. line-by-line annotations (Disease, Chemical)	F1 85.30

Decode API

The decode api can be called in the following way:

from lightner import decoder_wrapper
model = decoder_wrapper()
model.decode(["Ronaldo", "won", "'t", "score", "more", "than", "30", "goals", "for", "Juve", "."])

The decode() method also can conduct decoding at document level (takes list of list of str as input) or corpus level (takes list of list of list of str as input).

The decoder_wrapper method can be customized by choosing a different pre-trained model or passing an additional configs file as:

model = decoder_wrapper(URL_OR_PATH_TO_CHECKPOINT, configs)

And you can access the config options by:

lightner decode -h

Console

After installing and downloading the pre-trained mdoels, conduct the inference by

lightner decode -m MODEL_FILE -i INPUT_FILE -o OUTPUT_FILE

You can find more options by:

lightner decode -h

The current accepted paper format is as below (tokenized by line break and -DOCSTART- is optional):

-DOCSTART-

Ronaldo
won
't
score
more
30
goals
for
Juve
.

The output would be:

<PER> Ronaldo </PER> won 't score more than 30 goals for <ORG> Juve </ORG> .