MartinXPN / SpellNN

Multi-language Neural Spell-checking in TensorFlow 2

SpellNN

Multi-language Neural Spell-checking

Code organization

TODO

Datasets

TODO

python3 -m wiki2text.py --locale en --path datasets/en-articles.bz2 \                      # download latest pages-articles in XML format
             -o datasets/data/wiki/ --processes 4 datasets/en-articles.bz2 \    # parse the XML with WikiExtractor
             --wiki_dir datasets/data/wiki --output_path datasets/en_wiki.txt   # convert to .txt

# Cleanup
rm -rf datasets/data datasets/*articles.bz2

Model Architecture

TODO

Training

TODO

PYTHONHASHSEED=0 python3 -m spellnn.train \
                    construct_dataset --path datasets/en_wiki.txt --locale en - \
                    create_model Seq2SeqAttentionCNN --embedding_size 16 - \
                    train --epochs 50

Evaluation & Results

TODO

Installation

git clone [e-mail]:MartinXPN/SpellNN.git && cd SpellNN
pip install .[tf-gpu]  # Or `pip install .[tf]` if no gpu support is needed