File size: 6,459 Bytes
14cb0d3
 
 
 
 
 
 
 
 
 
 
c69a98a
dc1aacf
14cb0d3
 
 
 
 
 
 
 
 
 
 
dc1aacf
 
f6001fb
 
14cb0d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c5979c4
75f9996
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
408baf5
75f9996
 
0ccaaf6
 
408baf5
0ccaaf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
800ec67
 
 
0ccaaf6
 
b4ea28d
 
e676d8f
b4ea28d
408baf5
0ccaaf6
408baf5
 
 
0ccaaf6
 
 
408baf5
0ccaaf6
 
 
 
 
408baf5
0ccaaf6
408baf5
0ccaaf6
408baf5
0ccaaf6
408baf5
 
0ccaaf6
 
 
408baf5
 
 
 
0ccaaf6
408baf5
 
 
 
 
 
0ccaaf6
408baf5
0ccaaf6
408baf5
 
 
 
0ccaaf6
 
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
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_API_KEY = os.getenv('AI71_API_KEY')
# print(AI71_API_KEY)
huggingface_token = os.getenv('hf_token')
print("hf-token:",huggingface_token)

# 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]['Doctor']

    therapy_session_conversation.append(["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")
    print(f"therapy_session_conversation: {therapy_session_conversation}")
    return '', therapy_session_conversation, emotions_msg

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load model once globally for reuse
tokenizer = AutoTokenizer.from_pretrained("tiiuae/Falcon-H1-3B-Instruct")
use_cuda = torch.cuda.is_available()
model = AutoModelForCausalLM.from_pretrained(
    "tiiuae/Falcon-H1-1.5B-Deep-Instruct",
    torch_dtype=torch.float16 if use_cuda else torch.float32,
    device_map="auto" if use_cuda else None
)
def generate_falcon_response(prompt, max_new_tokens=300):
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1800).to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=0.7,
        top_p=0.9
    )
    decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return decoded_output[len(prompt):].strip()


def summarize_and_recommend(therapy_session_conversation):

    session_time = str(datetime.now().strftime("%Y-%m-%d %H:%M:%S"))
    # session_conversation = [item[0] for item in therapy_session_conversation]
    # session_conversation = [x for x in session_conversation if x is not None]
    session_conversation = [x for item in therapy_session_conversation for x in item if x is not None]

    session_conversation.insert(0, "Session_time: " + session_time)
    
    session_conversation_processed = '\n'.join(session_conversation)
    print("session_conversation_processed:", session_conversation_processed)

    # Summarization prompt
    summary_prompt = f"""You are an Expert Cognitive Behavioural Therapist and Precis writer.
Summarize STRICTLY the following session into concise, ethical, and clinically meaningful content.

Session:
{session_conversation_processed}

Format your response as:
Session Time:
Summary of the patient messages: 
Summary of therapist messages: 
Summary of the whole session: 
Ensure the entire summary is less than 300 tokens."""

    full_summary = generate_falcon_response(summary_prompt, max_new_tokens=300)
    print("\nFull summary:", full_summary)

    # Recommendation prompt
    recommendation_prompt = f"""You are an expert Cognitive Behavioural Therapist.
Based STRICTLY on the following summary, provide a clinically valid action plan for the patient.

Summary:
{full_summary}

Use this format:
- The patient is referred to...
- The patient is advised to...
- The patient is refrained from...
- It is suggested that the patient...
- Scheduled a follow-up session with the patient...

Ensure the list contains NOT MORE THAN 7 points and is in passive voice with proper tense."""

    full_recommendations = generate_falcon_response(recommendation_prompt, max_new_tokens=400)
    print("\nFull recommendations:", full_recommendations)

    chatbox = []
    return full_summary, full_recommendations, chatbox