Word Autocompletion Algorithm based on character-based LSTM

Character-based LSTM

• idea from char-rnn (https://github.com/karpathy/char-rnn), some implementation from sherjilozair's tensorflow char-rnn implementation (https://github.com/sherjilozair/char-rnn-tensorflow)
• Korean language model (jamo based), beam search for k-top result
• Tested on MAC OS X, Ubuntu 14.04

Demo for autocompletion

• prerequisites

# install requirements (python, numpy, tensorflow, java, hangul-utils etc ...)
cd data
./download.sh # downloading dataset

• Train text generative model

python train.py

• Test sentence, word sampling

# Currently using pretrained models, changed hard-coded directories to use other model
python sample.py

• Run test web server

# install requirements (flask)
python main.py
open localhost:5000