DeepSegment: A sentence segmenter that actually works!
DeepSegment is available as a free to use API (https://fastdeploy.notai.tech/free_apis) and as a self-hostable service via https://github.com/notAI-tech/fastDeploy
Note: For the original implementation please use the "master" branch of this repo.
Code documentation available at http://bpraneeth.com/docs
Installation:
# Tested with (keras==2.3.1; tensorflow==2.2.0) and (keras==2.2.4; tensorflow==1.14.0)
pip install --upgrade deepsegmentSupported languages:
en - english (Trained on data from various sources)
fr - french (Only Tatoeba data)
it - italian (Only Tatoeba data)
Usage:
from deepsegment import DeepSegment
# The default language is 'en'
segmenter = DeepSegment('en')
segmenter.segment('I am Batman i live in gotham')
# ['I am Batman', 'i live in gotham']Using with tf serving docker image
docker pull bedapudi6788/deepsegment_en:v2
docker run -d -p 8500:8500 bedapudi6788/deepsegment_en:v2from deepsegment import DeepSegment
# The default language is 'en'
segmenter = DeepSegment('en', tf_serving=True)
segmenter.segment('I am Batman i live in gotham')
# ['I am Batman', 'i live in gotham']Finetuning DeepSegment
Since one-size will never fit all, finetuning deepsegment's default models with your own data is encouraged.
from deepsegment import finetune, generate_data
x, y = generate_data(['my name', 'is batman', 'who are', 'you'], n_examples=10000)
vx, vy = generate_data(['my name', 'is batman'])
# NOTE: name, epochs, batch_size, lr are optional arguments.
finetune('en', x, y, vx, vy, name='finetuned_model_name', epochs=number_of_epochs, batch_size=batch_size, lr=learning_rate)Using with a finetuned checkpoint
from deepsegment import DeepSegment
segmenter = DeepSegment('en', checkpoint_name='finetuned_model_name')Training deepsegment on custom data: https://colab.research.google.com/drive/1CjYbdbDHX1UmIyvn7nDW2ClQPnnNeA_m
Similar Projects:
https://github.com/bminixhofer/nnsplit (with bindings for Python, Rust and Javascript.)