Phantom Pen - Web Application

These files contain the code for the Flask web application Phantom Pen. The model training (for GPT-2), web-scraping, and data cleaning codes are not included in this repository

Code Files

The code files have the following functions

• Text Files: Files for various corpora
  • gutenberg_texts
  • short_stories
  • Atlantic_subj_essays
  • concatenated_subj_files
• Training Code
  • train.py Slightly modified version of Nsheppherd's GPT2 training code
  • gpt2_training_clean.ipynb Code used to train the various models for the program
• Scraping texts and concatenating
  • web_scraping_atlantic_subjs.py Code used to scrape atlantic articles
  • web_scraping_short_stories.py Code used to scrape short stories
  • essay_concatenation_clean.py Code used to concatenate the articles, short stories, books
• Text generation and classification
  • text_generation_function.ipynb Code for the simple generate, classify & generate, and classify, extract, & generate functions of the application
  • classification_comparison.ipynb Comparison of various classifiers for the classify function
• extractive_text_summarizer folder: Code for explored extractive summarization implementations
  • text_summarizer_1_word_ct_cos_sim.ipynb: Extractive summary using word-count vectors and cosing similarity [Ultimately the chosen one]
  • text_summarizer_2_word_freq.ipynb: Extractive summary using word-frequency vectors
  • text_summarizer_3_word_embed.ipynb: Extractive summary using word embeddings
• writrly folder: Code for Flask application
  • HTML/CSS/Image Files
  • Python File of application
• Additional Code for Flask application
  • needed for sample generation
    • encoder.py: needed to encode words
    • sample.py: needed to sample words randomly
    • model.py: contains the definition of the model
  • To run application
    • run.py
• Misc Files
  • recurrent_neural_network_atlantic.ipynb: Implementation of LSTM RNN to generate words for Atlantic Education

Authors

• Mobolaji Williams - mowillia

Acknowledgments

• NSheppherd for code for training
• OpenAI for pretrained model