File size: 3,495 Bytes
25abf8c 04dafd4 25abf8c 04dafd4 25abf8c 04dafd4 25abf8c 380d8a4 261d5bd d9adf30 04dafd4 25abf8c 04dafd4 25abf8c 380d8a4 25abf8c 380d8a4 25abf8c 380d8a4 25abf8c 380d8a4 25abf8c 380d8a4 |
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 |
import streamlit as st
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from transformers import GPT2Tokenizer, GPT2LMHeadModel
from sentence_transformers import SentenceTransformer, util
import torch
import gdown
import os
# Download the CSV file from Hugging Face Spaces
url = 'https://huggingface.co/datasets/HEHEBOIBOT/PharmEvoDiabetesData/raw/main/medical_data.csv'
excel_file_path = os.path.join(os.path.expanduser("~"), 'medical_data.csv')
gdown.download(url, excel_file_path, quiet=False)
# Read the CSV file
try:
medical_df = pd.read_csv(excel_file_path, encoding='utf-8')
except UnicodeDecodeError:
medical_df = pd.read_csv(excel_file_path, encoding='latin1')
# TF-IDF Vectorization
vectorizer = TfidfVectorizer(stop_words='english')
X_tfidf = vectorizer.fit_transform(medical_df.iloc[:, 0]) # Accessing first column by index
# Load pre-trained GPT-2 model and tokenizer
model_name = "sshleifer/tiny-gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
model = GPT2LMHeadModel.from_pretrained(model_name)
# Load pre-trained Sentence Transformer model
sbert_model_name = "paraphrase-MiniLM-L6-v2"
sbert_model = SentenceTransformer(sbert_model_name)
# Function to answer medical questions using a combination of TF-IDF, LLM, and semantic similarity
def get_medical_response(question, vectorizer, X_tfidf, model, tokenizer, sbert_model, medical_df):
# TF-IDF Cosine Similarity
question_vector = vectorizer.transform([question])
tfidf_similarities = cosine_similarity(question_vector, X_tfidf).flatten()
# Find the most similar question using semantic similarity
question_embedding = sbert_model.encode(question, convert_to_tensor=True)
similarities = util.pytorch_cos_sim(question_embedding, sbert_model.encode(medical_df.iloc[:, 0].tolist(), convert_to_tensor=True)).flatten()
max_sim_index = similarities.argmax().item()
# LLM response generation
input_text = "DiBot: " + medical_df.iloc[max_sim_index][0]
input_ids = tokenizer.encode(input_text, return_tensors="pt")
attention_mask = torch.ones(input_ids.shape, dtype=torch.long)
pad_token_id = tokenizer.eos_token_id
lm_output = model.generate(input_ids, max_length=150, num_return_sequences=1, no_repeat_ngram_size=2, attention_mask=attention_mask, pad_token_id=pad_token_id)
lm_generated_response = tokenizer.decode(lm_output[0], skip_special_tokens=True)
# Compare similarities and choose the best response
if tfidf_similarities.max() > 0.5:
tfidf_index = tfidf_similarities.argmax()
return medical_df.iloc[tfidf_index][1] # Assuming 'Answers' is in the second column (index 1)
else:
return lm_generated_response
# Streamlit UI
st.title("DiBot")
if "messages" not in st.session_state:
st.session_state.messages = []
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
user_input = st.chat_input("You:")
if user_input:
response = get_medical_response(user_input, vectorizer, X_tfidf, model, tokenizer, sbert_model, medical_df)
st.session_state.messages.append({"role": "user", "content": user_input})
st.session_state.messages.append({"role": "assistant", "content": response})
# Display the chat messages
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
|