File size: 6,514 Bytes
c34fe27
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
# -*- coding: utf-8 -*-
"""Untitled0.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/13kE5uGoL2gfzSwTJli-WZolqCNBZXxNV
"""

import tensorflow as tf
import numpy as np
import pandas as pd
import streamlit as st
import re
import os
import csv
from tqdm import tqdm
import faiss
from nltk.translate.bleu_score import sentence_bleu
from datetime import datetime

def decontractions(phrase):
    """decontracted takes text and convert contractions into natural form.
     ref: https://stackoverflow.com/questions/19790188/expanding-english-language-contractions-in-python/47091490#47091490"""
    # specific
    phrase = re.sub(r"won\'t", "will not", phrase)
    phrase = re.sub(r"can\'t", "can not", phrase)
    phrase = re.sub(r"won\’t", "will not", phrase)
    phrase = re.sub(r"can\’t", "can not", phrase)

    # general
    phrase = re.sub(r"n\'t", " not", phrase)
    phrase = re.sub(r"\'re", " are", phrase)
    phrase = re.sub(r"\'s", " is", phrase)
    phrase = re.sub(r"\'d", " would", phrase)
    phrase = re.sub(r"\'ll", " will", phrase)
    phrase = re.sub(r"\'t", " not", phrase)
    phrase = re.sub(r"\'ve", " have", phrase)
    phrase = re.sub(r"\'m", " am", phrase)

    phrase = re.sub(r"n\’t", " not", phrase)
    phrase = re.sub(r"\’re", " are", phrase)
    phrase = re.sub(r"\’s", " is", phrase)
    phrase = re.sub(r"\’d", " would", phrase)
    phrase = re.sub(r"\’ll", " will", phrase)
    phrase = re.sub(r"\’t", " not", phrase)
    phrase = re.sub(r"\’ve", " have", phrase)
    phrase = re.sub(r"\’m", " am", phrase)

    return phrase


def preprocess(text):
    # convert all the text into lower letters
    # remove the words betweent brakets ()
    # remove these characters: {'$', ')', '?', '"', '’', '.',  '°', '!', ';', '/', "'", '€', '%', ':', ',', '('}
    # replace these spl characters with space: '\u200b', '\xa0', '-', '/'
    
    text = text.lower()
    text = decontractions(text)
    text = re.sub('[$)\?"’.°!;\'€%:,(/]', '', text)
    text = re.sub('\u200b', ' ', text)
    text = re.sub('\xa0', ' ', text)
    text = re.sub('-', ' ', text)
    return text


#importing bert tokenizer and loading the trained question embedding extractor model

from transformers import AutoTokenizer, TFGPT2Model
@st.cache(allow_output_mutation=True)
def return_biobert_tokenizer_model():
    '''returns pretrained biobert tokenizer and question extractor model'''
    biobert_tokenizer = AutoTokenizer.from_pretrained("cambridgeltl/BioRedditBERT-uncased")
    question_extractor_model1=tf.keras.models.load_model('question_extractor_model_2_11')
    return biobert_tokenizer,question_extractor_model1


#importing gpt2 tokenizer and loading the trained gpt2 model
from transformers import GPT2Tokenizer,TFGPT2LMHeadModel
@st.cache(allow_output_mutation=True)
def return_gpt2_tokenizer_model():
    '''returns pretrained gpt2 tokenizer and gpt2 model'''
    gpt2_tokenizer=GPT2Tokenizer.from_pretrained("gpt2")
    tf_gpt2_model=TFGPT2LMHeadModel.from_pretrained("tf_gpt2_model_2_118_50000")
    return gpt2_tokenizer,tf_gpt2_model

#preparing the faiss search
qa=pd.read_pickle('train_gpt_data.pkl')
question_bert = qa["Q_FFNN_embeds"].tolist()
answer_bert = qa["A_FFNN_embeds"].tolist()
question_bert = np.array(question_bert)
answer_bert = np.array(answer_bert)

question_bert = question_bert.astype('float32')
answer_bert = answer_bert.astype('float32')

answer_index = faiss.IndexFlatIP(answer_bert.shape[-1])

question_index = faiss.IndexFlatIP(question_bert.shape[-1])
answer_index.add(answer_bert)
question_index.add(question_bert)


print('finished initializing')

#defining function to prepare the data for gpt inference
#https://github.com/ash3n/DocProduct

def preparing_gpt_inference_data(gpt2_tokenizer,question,question_embedding):
  topk=20
  scores,indices=answer_index.search(
                  question_embedding.astype('float32'), topk)
  q_sub=qa.iloc[indices.reshape(20)]
  
  line = '`QUESTION: %s `ANSWER: ' % (
                        question)
  encoded_len=len(gpt2_tokenizer.encode(line))
  for i in q_sub.iterrows():
    line='`QUESTION: %s `ANSWER: %s ' % (i[1]['question'],i[1]['answer']) + line
    line=line.replace('\n','')
    encoded_len=len(gpt2_tokenizer.encode(line))
    if encoded_len>=1024:
      break
  return gpt2_tokenizer.encode(line)[-1024:]



#function to generate answer given a question and the required answer length

def give_answer(question,answer_len):
  preprocessed_question=preprocess(question)
  question_len=len(preprocessed_question.split(' '))
  truncated_question=preprocessed_question
  if question_len>500:
    truncated_question=' '.join(preprocessed_question.split(' ')[:500])
  biobert_tokenizer,question_extractor_model1= return_biobert_tokenizer_model()
  gpt2_tokenizer,tf_gpt2_model= return_gpt2_tokenizer_model()
  encoded_question= biobert_tokenizer.encode(truncated_question)
  max_length=512
  padded_question=tf.keras.preprocessing.sequence.pad_sequences(
      [encoded_question], maxlen=max_length, padding='post')
  question_mask=[[1 if token!=0 else 0 for token in question] for question in padded_question]
  embeddings=question_extractor_model1({'question':np.array(padded_question),'question_mask':np.array(question_mask)})
  gpt_input=preparing_gpt_inference_data(gpt2_tokenizer,truncated_question,embeddings.numpy())
  mask_start = len(gpt_input) - list(gpt_input[::-1]).index(4600) + 1
  input=gpt_input[:mask_start+1]
  if len(input)>(1024-answer_len):
   input=input[-(1024-answer_len):]
  gpt2_output=gpt2_tokenizer.decode(tf_gpt2_model.generate(input_ids=tf.constant([np.array(input)]),max_length=1024,temperature=0.7)[0])
  answer=gpt2_output.rindex('`ANSWER: ')
  return gpt2_output[answer+len('`ANSWER: '):]
  


#defining the final function to generate answer assuming default answer length to be 20
def final_func_1(question):
  answer_len=25
  return give_answer(question,answer_len)


def main():
    st.title('Medical Chatbot')
    question=st.text_input('Question',"Type Here")
    result=""
    if st.button('ask'):
        #with st.spinner("You Know! an apple a day keeps doctor away!"):
        start=datetime.now()
        result=final_func_1(question)
        end_time =datetime.now()
        st.success("Here is the answer")
        st.text(result)
        st.text("result recieved within "+str((end_time-start).total_seconds()))
            




if __name__=='__main__':
    main()