Update app.py

app.py

@@ -1,13 +1,10 @@
-import easyocr
-import numpy as np
-
-# create a reader once
-ocr_reader = easyocr.Reader(["en"], gpu=False)
 import gradio as gr
 import torch
 from transformers import pipeline as hf_pipeline, AutoModelForSequenceClassification, AutoTokenizer
 from PIL import Image
 import io
+import easyocr
+import numpy as np
 
 # ——— 1) Emotion Pipeline ————————————————————————————————————————————————
 emotion_pipeline = hf_pipeline(
@@ -18,11 +15,7 @@ emotion_pipeline = hf_pipeline(
 )
 
 def get_emotion_profile(text):
-    """
-    Returns a dict of emotion scores for the input text.
-    """
     results = emotion_pipeline(text)
-    # Some pipelines return a list of lists
     if isinstance(results, list) and isinstance(results[0], list):
         results = results[0]
     return {r["label"].lower(): round(r["score"], 3) for r in results}
@@ -55,11 +48,15 @@ THRESHOLDS = {
     "threat": 0.15
 }
 
-# ——— 3) Emotional-Tone Tagging —————————————————————————————————————————————
+# ——— 3) Initialize EasyOCR reader ————————————————————————————————————————————
+ocr_reader = easyocr.Reader(["en"], gpu=False)
+
+# ——— 4) Emotional-Tone Tagging —————————————————————————————————————————————
 def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
     """
     Assigns one of 18 nuanced tone categories based on emotion scores, patterns, and text.
     """
+    # unpack all emotion scores before any rules
     sadness  = emotion_profile.get("sadness",  0)
     joy      = emotion_profile.get("joy",      0)
     neutral  = emotion_profile.get("neutral",  0)
@@ -68,62 +65,40 @@ def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
     fear     = emotion_profile.get("fear",     0)
     surprise = emotion_profile.get("surprise", 0)
 
-def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
-    if "support" in text_lower or "hope" in text_lower or "grace" in text_lower:
+    # 0. Support override
+    if any(k in text_lower for k in ["support", "hope", "grace"]):
         return "supportive"
-    if (
-        sadness > 0.4 and
-        any(p in patterns for p in ["blame shifting", "guilt tripping", "recovery phase"])
-    ):
+
+    # 1. Performative Regret
+    if sadness > 0.4 and any(p in patterns for p in ["blame shifting", "guilt tripping", "recovery phase"]):
         return "performative regret"
 
     # 2. Coercive Warmth
-    if (
-        (joy > 0.3 or sadness > 0.4) and
-        any(p in patterns for p in ["control", "gaslighting"])
-    ):
+    if (joy > 0.3 or sadness > 0.4) and any(p in patterns for p in ["control", "gaslighting"]):
         return "coercive warmth"
 
     # 3. Cold Invalidation
-    if (
-        (neutral + disgust) > 0.5 and
-        any(p in patterns for p in ["dismissiveness", "projection", "obscure language"])
-    ):
+    if (neutral + disgust) > 0.5 and any(p in patterns for p in ["dismissiveness", "projection", "obscure language"]):
         return "cold invalidation"
 
     # 4. Genuine Vulnerability
-    if (
-        (sadness + fear) > 0.5 and
-        all(p == "recovery phase" for p in patterns)
-    ):
+    if (sadness + fear) > 0.5 and all(p == "recovery phase" for p in patterns):
         return "genuine vulnerability"
 
     # 5. Emotional Threat
-    if (
-        (anger + disgust) > 0.5 and
-        any(p in patterns for p in ["control", "threat", "insults", "dismissiveness"])
-    ):
+    if (anger + disgust) > 0.5 and any(p in patterns for p in ["control", "threat", "insults", "dismissiveness"]):
         return "emotional threat"
 
     # 6. Weaponized Sadness
-    if (
-        sadness > 0.6 and
-        any(p in patterns for p in ["guilt tripping", "projection"])
-    ):
+    if sadness > 0.6 and any(p in patterns for p in ["guilt tripping", "projection"]):
         return "weaponized sadness"
 
     # 7. Toxic Resignation
-    if (
-        neutral > 0.5 and
-        any(p in patterns for p in ["dismissiveness", "obscure language"])
-    ):
+    if neutral > 0.5 and any(p in patterns for p in ["dismissiveness", "obscure language"]):
         return "toxic resignation"
 
     # 8. Indignant Reproach
-    if (
-        anger > 0.5 and
-        any(p in patterns for p in ["guilt tripping", "contradictory statements"])
-    ):
+    if anger > 0.5 and any(p in patterns for p in ["guilt tripping", "contradictory statements"]):
         return "indignant reproach"
 
     # 9. Confrontational
@@ -131,10 +106,7 @@ def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
         return "confrontational"
 
     # 10. Passive Aggression
-    if (
-        neutral > 0.6 and
-        any(p in patterns for p in ["dismissiveness", "projection"])
-    ):
+    if neutral > 0.6 and any(p in patterns for p in ["dismissiveness", "projection"]):
         return "passive aggression"
 
     # 11. Sarcastic Mockery
@@ -146,11 +118,7 @@ def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
         return "menacing threat"
 
     # 13. Pleading Concern
-    if (
-        sadness > 0.3 and
-        any(k in text_lower for k in APOLOGY_KEYWORDS) and
-        not patterns
-    ):
+    if sadness > 0.3 and any(k in text_lower for k in APOLOGY_KEYWORDS) and not patterns:
         return "pleading concern"
 
     # 14. Fear-mongering
@@ -175,43 +143,30 @@ def get_emotional_tone_tag(emotion_profile, patterns, text_lower):
 
     return None
 
-# ——— 4) Single-message Analysis ———————————————————————————————————————————
+# ——— 5) Single message analysis ———————————————————————————————————————————
 def analyze_message(text):
     text_lower = text.lower()
-    # 1) Emotion
     emotion_profile = get_emotion_profile(text)
-    # 2) Patterns
     toks = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
         logits = model(**toks).logits.squeeze(0)
     scores = torch.sigmoid(logits).cpu().numpy()
-    active_patterns = [lab for lab, sc in zip(LABELS, scores) if sc >= THRESHOLDS[lab]]
-    # append recovery-phase if apology
+    active_patterns = [label for label, prob in zip(LABELS, scores) if prob >= THRESHOLDS[label]]
     if any(k in text_lower for k in APOLOGY_KEYWORDS) and "recovery phase" not in active_patterns:
         active_patterns.append("recovery phase")
-    # 3) Tone
     tone_tag = get_emotional_tone_tag(emotion_profile, active_patterns, text_lower)
-    return {
-        "emotion_profile": emotion_profile,
-        "active_patterns": active_patterns,
-        "tone_tag": tone_tag
-    }
+    return {"emotion_profile": emotion_profile, "active_patterns": active_patterns, "tone_tag": tone_tag}
 
-# ——— 5) Composite Wrapper ————————————————————————————————————————————————
+# ——— 6) Composite wrapper ———————————————————————————————————————————————
 def analyze_composite(uploaded_file, *texts):
     outputs = []
-
-    # 1) Handle an uploaded file
     if uploaded_file is not None:
-        # uploaded_file may be a file-like object or just a path string
         try:
             raw = uploaded_file.read()
         except Exception:
-            # fall back to opening the path
             with open(uploaded_file, "rb") as f:
                 raw = f.read()
 
-        # get the filename (or just lowercase the string if it's a path)
         name = (
             uploaded_file.name.lower() if hasattr(uploaded_file, "name") else uploaded_file.lower()
         )
@@ -249,19 +204,13 @@ def analyze_composite(uploaded_file, *texts):
 
 # ——— 7) Gradio interface ———————————————————————————————————————————————
 message_inputs = [gr.Textbox(label="Message")]
-
 iface = gr.Interface(
     fn=analyze_composite,
-    inputs=[
-        gr.File(
-            file_types=[".txt", ".png", ".jpg", ".jpeg"],
-            label="Upload text or image"
-        )
-    ] + message_inputs,   # <— here the list closes, then we concatenate
+    inputs=[gr.File(file_types=[".txt",".png",".jpg",".jpeg"], label="Upload text or image")] + message_inputs,
     outputs=gr.Textbox(label="Analysis"),
     title="Tether Analyzer (extended tone tags)",
     description="Emotion profiling, pattern tags, and a wide set of nuanced tone categories—no abuse score or DARVO."
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    iface.launch()