import os

os.system("pip install -U transformers==3.0.0")
os.system("pip install nltk torch docx2txt")
os.system("python -m nltk.downloader punkt")

import gradio as gr
import pandas as pd
from question_generation.pipelines import pipeline
import docx2txt

qa_list = []
def process_file(Notes):

  nlp = pipeline("question-generation", model="valhalla/t5-small-qg-prepend", qg_format="prepend")
  
  target_word_doc = Notes.name
  raw_word_file = docx2txt.process(target_word_doc)

  #remove empty lines
  preprocessed_sentence_list = [i for i in raw_word_file.splitlines() if i != ""]

  #grab content
  processed_sentence_list = []
  content = False
  for i in preprocessed_sentence_list:
    if "Outline" in i:
      content = True
      continue
    if "Summary Learning Points" in i:
      content = False
      continue
    if "Learning Activity" in i:
      content = False
      continue
    if content == True:
      processed_sentence_list.append(i.lstrip())

  qa_list.extend(nlp(" ".join(processed_sentence_list)))
  formatted_questions = "\n".join([str(idx+1) + ". " + i["question"] for idx, i in enumerate(qa_list)])
  formatted_answers = "\n".join([str(idx+1) + ". " + i["answer"] for idx, i in enumerate(qa_list)])
  return [formatted_questions, formatted_answers]

def reveal_answer():
  global qa_list
  
  qa_list = []
  return formatted_answers

io = gr.Interface(process_file, "file", outputs=
                  [gr.Textbox(lines=1, label="Questions"), 
                   gr.Textbox(lines=1, label="Answers")])
io.launch()