Goal

Design and implement an auto-complete server using scripts from all the episodes of Seinfeld. The server will try to predict how Jerry Seinfeld would finish a given incomplete sentence.

Dataset obtained from https://www.kaggle.com/thec03u5/seinfeld-chronicles

Instructions

Git clone the code

git clone https://github.com/ArbinTimilsina/AutoCompleteJerry.git
cd AutoCompleteJerry

To train the model, do

python train_model.py --help

# Example
python train_model.py -e 1

To use the model, do

python run_server.py

and then curl the following (or such) in a new terminal (or paste into a web browser)

http://localhost:5050/autocomplete?seed=What+is+the+deal

You will get output similar to

Replace the string after ?seed= to change the seed and see suggested completions!

Additional information

To create a conda environment (Python 3)

conda create --name envAutoCompleteJerry python=3.5
conda activate envAutoCompleteJerry
pip install --upgrade pip
pip install -r requirements/cpu_requirements.txt
conda install pydot graphviz

Download the GloVe word embeddings

Head to https://nlp.stanford.edu/projects/glove/ and download the pre-computed embeddings from 2014 English Wikipedia. Un-zip it.

wget -O glove.6B.zip http://nlp.stanford.edu/data/glove.6B.zip
unzip glove.6B.zip -d glove
mv glove/* glove.6B

Get spaCy

pip install spacy
python -m spacy download en_core_web_sm

To switch Keras backend to TensorFlow

KERAS_BACKEND=tensorflow python -c "from keras import backend"

To open jupyter notebook

# Create an IPython kernel for the environment
python -m ipykernel install --user --name envAutoCompleteJerry --display-name "envAutoCompleteJerry"

# Open the notebook
jupyter notebook miscellaneous/model_creation_playground.ipynb

# Make sure to change the kernel to envAutoComplete using the drop-down menu (Kernel > Change kernel > envAutoCompleteJerry)