app.py

# -*- coding: utf-8 -*-
"""assessment3_Mark_Hayden.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1gTZtBnC7kvWELzlIiTOkRd1aJaeXxJtm

Hello professor, I was able to deploy the app to HuggingFace but ran into an issue that I think is on HuggingFace/Gradio side: /usr/local/lib/python3.10/site-packages/gradio/blocks.py:626: UserWarning: Cannot load dark-grass. Caught Exception: 404 Client Error: Not Found for url: https://huggingface.co/api/spaces/dark-grass (Request ID: Root=1-661c106f-039767d826b7f84a6a7b4ac4;63fe1a3b-5196-4853-925d-fe96a80c1756).

There is documentation on this that I can see, and the interface runs in the notebook so I don't believe it's a code/library issue but I left the Space up so you can see.

Gradio Space: https://huggingface.co/spaces/MHayden/enron_qa

**Part 1:**
This section of the notebook is used for pulling the Enron dataset down from Kaggle.
It contains the preprocessing steps I used. I have saved the output as a pickle file for Part 2 below.
"""

import pandas as pd
import numpy as np
import pickle
from transformers import pipeline
import gradio as gr

with open('file.pkl', 'rb') as file:
    # Call load method to deserialze
    df = pickle.load(file)

# QA pipeline setup
from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering
model = TFAutoModelForQuestionAnswering.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad",return_dict=False)
tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
nlp = pipeline("question-answering", model=model, tokenizer=tokenizer)

df['bankrupt'] = df['emails'].str.lower().str.contains(r'\bbankrupt\b')
df['litigation'] = df['emails'].str.lower().str.contains(r'\blitigation\b')
df['fraud'] = df['emails'].str.lower().str.contains(r'\bfraud\b')
df['talking_points'] = df['emails'].str.lower().str.contains(r'\btalking\spoint\b')
df['shutdown'] = df['emails'].str.lower().str.contains(r'\bshutdown\b')

# Creating dataframe for Gradio
gradio_df = df.loc[(df['bankrupt'] == True) | (df['litigation'] == True) | (df['fraud'] == True) | (df['talking_points'] == True)
                  | (df['shutdown'] == True)]

example_questions = {}
tags = {}
for i in range(gradio_df.shape[0]):
    example_questions['example_' + str(i+1)] = gradio_df['emails'].iloc[i]
    tags['tag' + str(i+1)] = gradio_df[['bankrupt', 'litigation', 'fraud', 'talking_points','shutdown']].iloc[i]

import gradio as gr
# creating the function
def func(context, question,tags):
  result = nlp(question = question, context=context,tags = tags)
  return result['answer']

question =  'Search for a keyword/phrase using Tags to guide you'

# creating the interface
app = gr.Interface(fn=func, inputs = ['textbox', 'text','text'], outputs = 'textbox',
                   title = 'Question Answering bot',
                   description = 'Please choose an example below. You can write a question and recieve an answer, please use tags to guide you.',
                   examples = [[example_questions['example_1'],question,tags['tag1']],[example_questions['example_2'],question,tags['tag2']],[example_questions['example_3'],question,tags['tag3']],[example_questions['example_4'],question,tags['tag4']],[example_questions['example_5'],question,tags['tag5']],[example_questions['example_6'],question,tags['tag6']],[example_questions['example_7'],question,tags['tag7']],[example_questions['example_8'],question,tags['tag8']],[example_questions['example_9'],question,tags['tag9']],[example_questions['example_10'],question,tags['tag10']],[example_questions['example_11'],question,tags['tag11']],[example_questions['example_12'],question,tags['tag12']],[example_questions['example_13'],question,tags['tag13']],[example_questions['example_14'],question,tags['tag14']],[example_questions['example_15'],question,tags['tag15']],[example_questions['example_16'],question,tags['tag16']],[example_questions['example_17'],question,tags['tag17']],[example_questions['example_18'],question,tags['tag18']],[example_questions['example_19'],question,tags['tag19']],[example_questions['example_20'],question,tags['tag20']],[example_questions['example_21'],question,tags['tag21']],[example_questions['example_22'],question,tags['tag22']],[example_questions['example_23'],question,tags['tag23']],[example_questions['example_24'],question,tags['tag24']],[example_questions['example_25'],question,tags['tag25']],[example_questions['example_26'],question,tags['tag26']],[example_questions['example_27'],question,tags['tag27']],[example_questions['example_28'],question,tags['tag28']],[example_questions['example_29'],question,tags['tag29']],[example_questions['example_30'],question,tags['tag30']],[example_questions['example_31'],question,tags['tag31']],[example_questions['example_32'],question,tags['tag32']],[example_questions['example_33'],question,tags['tag33']],[example_questions['example_34'],question,tags['tag34']],[example_questions['example_35'],question,tags['tag35']],[example_questions['example_36'],question,tags['tag36']],[example_questions['example_37'],question,tags['tag37']],[example_questions['example_38'],question,tags['tag38']],[example_questions['example_39'],question,tags['tag39']],[example_questions['example_40'],question,tags['tag40']],[example_questions['example_41'],question,tags['tag41']],[example_questions['example_42'],question,tags['tag42']],[example_questions['example_43'],question,tags['tag43']],[example_questions['example_44'],question,tags['tag44']],[example_questions['example_45'],question,tags['tag45']],[example_questions['example_46'],question,tags['tag46']],[example_questions['example_47'],question,tags['tag47']],[example_questions['example_48'],question,tags['tag48']],[example_questions['example_49'],question,tags['tag49']],[example_questions['example_50'],question,tags['tag50']],[example_questions['example_51'],question,tags['tag51']],[example_questions['example_52'],question,tags['tag52']],[example_questions['example_53'],question,tags['tag53']],[example_questions['example_54'],question,tags['tag54']],[example_questions['example_55'],question,tags['tag55']],[example_questions['example_56'],question,tags['tag56']],[example_questions['example_57'],question,tags['tag57']],[example_questions['example_58'],question,tags['tag58']],[example_questions['example_59'],question,tags['tag59']],[example_questions['example_60'],question,tags['tag60']],[example_questions['example_61'],question,tags['tag61']],[example_questions['example_62'],question,tags['tag62']],[example_questions['example_63'],question,tags['tag63']],[example_questions['example_64'],question,tags['tag64']],[example_questions['example_65'],question,tags['tag65']],[example_questions['example_66'],question,tags['tag66']],[example_questions['example_67'],question,tags['tag67']],[example_questions['example_68'],question,tags['tag68']],[example_questions['example_69'],question,tags['tag69']],[example_questions['example_70'],question,tags['tag70']],[example_questions['example_71'],question,tags['tag71']],[example_questions['example_72'],question,tags['tag72']],[example_questions['example_73'],question,tags['tag73']],[example_questions['example_74'],question,tags['tag74']],[example_questions['example_75'],question,tags['tag75']],[example_questions['example_76'],question,tags['tag76']],[example_questions['example_77'],question,tags['tag77']],[example_questions['example_78'],question,tags['tag78']],[example_questions['example_79'],question,tags['tag79']],[example_questions['example_80'],question,tags['tag80']],[example_questions['example_81'],question,tags['tag81']]]
                   )

# launching the app
app.launch(debug=True)

app.close()