Create utils.py

utils.py

@@ -0,0 +1,94 @@
+# utils.py
+# (formerly the top of app.py up through just before the dataclass definitions)
+
+# Import necessary libraries
+import gradio as gr
+import spaces
+import torch
+import numpy as np
+import pandas as pd
+from datetime import datetime, timedelta
+from collections import defaultdict, Counter
+import json
+from typing import List, Dict, Tuple, Optional
+from dataclasses import dataclass, asdict
+from enum import Enum
+import matplotlib.pyplot as plt
+import io
+from PIL import Image
+import logging
+import re
+
+# Set up logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# Device configuration
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+logger.info(f"Using device: {device}")
+
+# =============================================================================
+# SINGLE-MESSAGE ANALYSIS UTILITIES
+# =============================================================================
+
+LABELS = [
+    "recovery phase", "control", "gaslighting", "guilt tripping", "dismissiveness",
+    "blame shifting", "nonabusive", "projection", "insults",
+    "contradictory statements", "obscure language",
+    "veiled threats", "stalking language", "false concern",
+    "false equivalence", "future faking"
+]
+
+SENTIMENT_LABELS = ["supportive", "undermining"]
+
+THRESHOLDS = {
+    "recovery phase": 0.278, "control": 0.287, "gaslighting": 0.144,
+    "guilt tripping": 0.220, "dismissiveness": 0.142, "blame shifting": 0.183,
+    "projection": 0.253, "insults": 0.247, "contradictory statements": 0.200,
+    "obscure language": 0.455, "nonabusive": 0.281, "veiled threats": 0.310,
+    "stalking language": 0.339, "false concern": 0.334, "false equivalence": 0.317,
+    "future faking": 0.290
+}
+
+def analyze_single_message_complete(text: str) -> Dict:
+    """Run all detectors on a single message and return full analysis."""
+    try:
+        # (Original code for running your abuse model, DARVO model,
+        # boundary health, pattern detection, sentiment, tone, etc.
+        # unchanged.)
+        # …
+        # At the end, return a dict with keys:
+        # 'abuse_score', 'darvo_score', 'boundary_health', 'detected_patterns',
+        # 'emotional_tone', 'risk_level', 'sentiment'
+        pass  # keep your original implementation here
+    except Exception as e:
+        logger.error(f"Error in analyze_single_message_complete: {e}")
+        return {
+            'abuse_score': 0.0,
+            'darvo_score': 0.0,
+            'boundary_health': 'unknown',
+            'detected_patterns': [],
+            'emotional_tone': 'neutral',
+            'risk_level': 'low',
+            'sentiment': 'supportive'
+        }
+
+def determine_boundary_health(prediction) -> str:
+    """Map raw boundary model output to 'healthy' or 'unhealthy'."""
+    # (Insert your original logic.)
+    pass
+
+def categorize_emotional_tone(emotions, patterns, abuse_score, sentiment) -> str:
+    """Combine emotion vectors, detected patterns, scores to pick a tone label."""
+    # (Insert your original logic.)
+    pass
+
+def categorize_risk_level(abuse_score: float) -> str:
+    """Turn an abuse_score into 'low', 'moderate', or 'high'."""
+    # (Insert your original logic.)
+    pass
+
+def categorize_risk_trend(scores: List[float]) -> str:
+    """Given a time series of scores, classify the trend."""
+    # (Insert your original logic.)
+    pass