Spaces:
Running
Running
File size: 11,228 Bytes
14cb0d3 c5979c4 75f9996 55acba8 75f9996 3422a89 b952f74 8e47518 f13eed9 995aa40 8afa4ac c5315f1 8afa4ac 8e47518 8afa4ac 55acba8 8afa4ac 55acba8 8afa4ac 55acba8 8afa4ac 55acba8 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 |
import torch
import numpy as np
import io
import matplotlib.pyplot as plt
import pandas as pd
from sentence_transformers import SentenceTransformer
from transformers import pipeline
from datetime import datetime
from PIL import Image
import os
from datetime import datetime
from openai import OpenAI
from ai71 import AI71
if torch.cuda.is_available():
model = model.to('cuda')
dials_embeddings = pd.read_pickle('https://huggingface.co/datasets/vsrinivas/CBT_dialogue_embed_ds/resolve/main/kaggle_therapy_embeddings.pkl')
with open ('emotion_group_labels.txt') as file:
emotion_group_labels = file.read().splitlines()
embed_model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
classifier = pipeline("zero-shot-classification", model ='facebook/bart-large-mnli')
AI71_BASE_URL = "https://api.ai71.ai/v1/"
AI71_API_KEY = os.getenv('AI71_API_KEY')
# Detect emotions from patient dialogues
def detect_emotions(text):
emotion = classifier(text, candidate_labels=emotion_group_labels, batch_size=16)
top_5_scores = [i/sum(emotion['scores'][:5]) for i in emotion['scores'][:5]]
top_5_emotions = emotion['labels'][:5]
emotion_set = {l: "{:.2%}".format(s) for l, s in zip(top_5_emotions, top_5_scores)}
return emotion_set
# Measure cosine similarity between a pair of vectors
def cosine_distance(vec1,vec2):
cosine = (np.dot(vec1, vec2)/(np.linalg.norm(vec1)*np.linalg.norm(vec2)))
return cosine
# Generate an image of trigger emotions
def generate_triggers_img(items):
labels = list(items.keys())
values = [float(v.strip('%')) for v in items.values()] # Convert to float for plotting
new_items = {k:v for k, v in zip(labels, values)}
new_items = dict(sorted(new_items.items(), key=lambda item: item[1]))
labels = list(new_items.keys())
values = list(new_items.values())
fig, ax = plt.subplots(figsize=(10, 6))
colors = plt.cm.viridis(np.linspace(0, 1, len(labels)))
bars = ax.barh(labels, values, color=colors)
for spine in ax.spines.values():
spine.set_visible(False)
ax.tick_params(axis='y', labelsize=18)
ax.xaxis.set_visible(False)
ax.yaxis.set_ticks_position('none')
for bar in bars:
width = bar.get_width()
ax.text(width, bar.get_y() + bar.get_height()/2, f'{width:.2f}%',
ha='left', va='center', fontweight='bold', fontsize=18)
plt.tight_layout()
plt.savefig('triggeres.png')
triggers_img = Image.open('triggeres.png')
return triggers_img
def get_doc_response_emotions(user_message, therapy_session_conversation):
user_messages = []
user_messages.append(user_message)
emotion_set = detect_emotions(user_message)
print(emotion_set)
emotions_msg = generate_triggers_img(emotion_set)
user_embedding = embed_model.encode(user_message, device='cuda' if torch.cuda.is_available() else 'cpu')
similarities =[]
for v in dials_embeddings['embeddings']:
similarities.append(cosine_distance(user_embedding,v))
top_match_index = similarities.index(max(similarities))
# doc_response = dials_embeddings.iloc[top_match_index+1]['Doctor']
doc_response = dials_embeddings.iloc[top_match_index]['Doctor']
therapy_session_conversation.append(["User: "+user_message, "Therapist: "+doc_response])
# session_conversation.extend(["User: "+user_message, "Therapist: "+doc_response])
print(f"User's message: {user_message}")
print(f"RAG Matching message: {dials_embeddings.iloc[top_match_index]['Patient']}")
print(f"Therapist's response: {dials_embeddings.iloc[top_match_index]['Doctor']}\n\n")
return '', therapy_session_conversation, emotions_msg
def summarize_and_recommend(therapy_session_conversation):
print("tcs:", therapy_session_conversation, type(therapy_session_conversation))
session_conversation = list(therapy_session_conversation.value)
# session_conversation = [item[0] for item in session_conversation]
print("Session conversation:", session_conversation)
session_time = str(datetime.now().strftime("%Y-%m-%d %H:%M:%S"))
session_conversation_processed = session_conversation.copy()
session_conversation_processed.insert(0, "Session_time: "+session_time)
session_conversation_processed ='\n'.join(session_conversation_processed)
print("session_conversation_processed:", session_conversation_processed)
full_summary = ""
for chunk in AI71(AI71_API_KEY).chat.completions.create(
model="tiiuae/falcon-180b-chat",
messages=[
{"role": "system", "content": """You are an Expert Cognitive Behavioural Therapist and Precis writer.
Summarize 'STRICTLY' the below user content <<<session_conversation_processed>>> 'ONLY' into useful, ethical, relevant and realistic phrases with a format
Session Time:
Summary of the patient messages: #in two to four sentences
Summary of therapist messages: #in two to three sentences:
Summary of the whole session: # in two to three sentences. Ensure the entire session summary strictly does not exceed 100 tokens."""},
{"role": "user", "content": session_conversation_processed},
],
stream=True,
):
if chunk.choices[0].delta.content:
summary = chunk.choices[0].delta.content
full_summary += summary
full_summary = full_summary.replace('User:', '').strip()
print("\n")
print("Full summary:", full_summary)
full_recommendations = ""
for chunk in AI71(AI71_API_KEY).chat.completions.create(
model="tiiuae/falcon-180b-chat",
messages=[
{"role": "system", "content": """You are an expert Cognitive Behavioural Therapist.
Based on 'STRICTLY' the full summary <<<full_summary>>> 'ONLY' provide clinically valid, useful, appropriate action plan for the Patient as a bullted list.
The list shall contain both medical and non medical prescriptions, dos and donts. The format of response shall be in passive voice with proper tense.
- The patient is referred to........ #in one sentence
- The patient is advised to ........ #in one sentence
- The patient is refrained from........ #in one sentence
- It is suggested that tha patient ........ #in one sentence
- Scheduled a follow-up session with the patient........#in one sentence
*Ensure the list contains NOT MORE THAN 7 points"""},
{"role": "user", "content": full_summary},
],
stream=True,
):
if chunk.choices[0].delta.content:
rec = chunk.choices[0].delta.content
full_recommendations += rec
full_recommendations = full_recommendations.replace('User:', '').strip()
print("\n")
print("Full recommendations:", full_recommendations)
therapy_session_conversation=[]
return full_summary, full_recommendations
# class process_session():
# def __init__(self):
# self.session_conversation=[]
# def get_doc_response_emotions(self, user_message, therapy_session_conversation):
# user_messages = []
# user_messages.append(user_message)
# emotion_set = detect_emotions(user_message)
# print(emotion_set)
# emotions_msg = generate_triggers_img(emotion_set)
# user_embedding = embed_model.encode(user_message, device='cuda' if torch.cuda.is_available() else 'cpu')
# similarities =[]
# for v in dials_embeddings['embeddings']:
# similarities.append(cosine_distance(user_embedding,v))
# top_match_index = similarities.index(max(similarities))
# # doc_response = dials_embeddings.iloc[top_match_index+1]['Doctor']
# doc_response = dials_embeddings.iloc[top_match_index]['Doctor']
# therapy_session_conversation.append(["User: "+user_message, "Therapist: "+doc_response])
# self.session_conversation.extend(["User: "+user_message, "Therapist: "+doc_response])
# print(f"User's message: {user_message}")
# print(f"RAG Matching message: {dials_embeddings.iloc[top_match_index]['Patient']}")
# print(f"Therapist's response: {dials_embeddings.iloc[top_match_index]['Doctor']}\n\n")
# return '', therapy_session_conversation, emotions_msg
# def summarize_and_recommend(self):
# session_time = str(datetime.now().strftime("%Y-%m-%d %H:%M:%S"))
# session_conversation_processed = self.session_conversation.copy()
# session_conversation_processed.insert(0, "Session_time: "+session_time)
# session_conversation_processed ='\n'.join(session_conversation_processed)
# print("Session conversation:", session_conversation_processed)
# full_summary = ""
# for chunk in AI71(AI71_API_KEY).chat.completions.create(
# model="tiiuae/falcon-180b-chat",
# messages=[
# {"role": "system", "content": """You are an Expert Cognitive Behavioural Therapist and Precis writer.
# Summarize the below user content <<<session_conversation_processed>>> into useful, ethical, relevant and realistic phrases with a format
# Session Time:
# Summary of the patient messages: #in two to four sentences
# Summary of therapist messages: #in two to three sentences:
# Summary of the whole session: # in two to three sentences. Ensure the entire session summary strictly does not exceed 100 tokens."""},
# {"role": "user", "content": session_conversation_processed},
# ],
# stream=True,
# ):
# if chunk.choices[0].delta.content:
# summary = chunk.choices[0].delta.content
# full_summary += summary
# full_summary = full_summary.replace('User:', '').strip()
# print("\n")
# print("Full summary:", full_summary)
# full_recommendations = ""
# for chunk in AI71(AI71_API_KEY).chat.completions.create(
# model="tiiuae/falcon-180b-chat",
# messages=[
# {"role": "system", "content": """You are an expert Cognitive Behavioural Therapist.
# Based on the full summary <<<full_summary>>> provide clinically valid, useful, appropriate action plan for the Patient as a bullted list.
# The list shall contain both medical and non medical prescriptions, dos and donts. The format of response shall be in passive voice with proper tense.
# - The patient is referred to........ #in one sentence
# - The patient is advised to ........ #in one sentence
# - The patient is refrained from........ #in one sentence
# - It is suggested that tha patient ........ #in one sentence
# - Scheduled a follow-up session with the patient........#in one sentence
# *Ensure the list contains NOT MORE THAN 7 points"""},
# {"role": "user", "content": full_summary},
# ],
# stream=True,
# ):
# if chunk.choices[0].delta.content:
# rec = chunk.choices[0].delta.content
# full_recommendations += rec
# full_recommendations = full_recommendations.replace('User:', '').strip()
# print("\n")
# print("Full recommendations:", full_recommendations)
# self.session_conversation=[]
# return full_summary, full_recommendations
|