Update app.py

app.py

@@ -1,3 +1,5 @@
+import os
+import json
 import streamlit as st
 import torch
 import torch.nn.functional as F
@@ -6,59 +8,15 @@ import re
 import math
 import logging
 
-# --- Heuristic weights & config (from your PHP) ---
-SIGMOID_K = 0.5
+# --- Load heuristic weights from environment secrets ---
+@st.cache_resource
+def load_heuristic_weights():
+    ai = json.loads(os.environ["AI_WEIGHTS_JSON"])
+    og = json.loads(os.environ["OG_WEIGHTS_JSON"])
+    return ai, og
 
-AI_WEIGHTS = {
-    'users': 1.0000, 'significant': 0.8061, 'despite': 0.5934, 'additionally': 0.5313,
-    'potential': 0.4900, 'features': 0.4824, 'various': 0.4785, 'regarding': 0.4689,
-    'remains': 0.4403, 'featuring': 0.4031, 'experience': 0.4026, 'including': 0.3963,
-    'challenges': 0.3548, 'allowing': 0.3528, 'enhance': 0.3437, 'aims': 0.3238,
-    'leading': 0.3064, 'user': 0.3054, 'recent': 0.2717, 'concerns': 0.2707,
-    'capabilities': 0.2684, 'technology': 0.2636, 'devices': 0.2616, 'following': 0.2551,
-    'anticipated': 0.2484, 'unique': 0.2418, 'expressed': 0.2403, 'innovative': 0.2383,
-    'design': 0.2379, 'remain': 0.2371, 'previous': 0.2331, 'priced': 0.2325,
-    'launch': 0.2324, 'enhancing': 0.2319, 'showcasing': 0.2305, 'feature': 0.2288,
-    'particularly': 0.2287, 'set': 0.2228, 'aimed': 0.2216, 'highlighted': 0.2192,
-    'ongoing': 0.2188, 'access': 0.2182, 'available': 0.2159, 'alongside': 0.2144,
-    'introduced': 0.2133, 'previously': 0.2122, 'highlighting': 0.2113, 'models': 0.2081,
-    'faced': 0.2057, 'platforms': 0.2055, 'updates': 0.2037, 'offers': 0.2032,
-    'significantly': 0.2027, 'issues': 0.2021, 'emphasized': 0.1977, 'initially': 0.1955,
-    'content': 0.1926, 'emphasizing': 0.1924, 'options': 0.1874, 'performance': 0.1864,
-    'initial': 0.1832, 'notable': 0.1821, 'additional': 0.1812, 'individuals': 0.1804,
-    'initiative': 0.1802, 'enhanced': 0.1797, 'release': 0.1797, 'currently': 0.1790,
-    'traditional': 0.1769, 'future': 0.1731, 'expected': 0.1725, 'applications': 0.1707,
-    'indicating': 0.1699, 'notably': 0.1658, 'insights': 0.1656, 'noted': 0.1645,
-    'players': 0.1645, 'narrative': 0.1642, 'landscape': 0.1640, 'upcoming': 0.1634,
-    'providing': 0.1631, 'offering': 0.1615, 'enabling': 0.1610, 'gaming': 0.1595,
-    'compared': 0.1553, 'indicated': 0.1539, 'extensive': 0.1530, 'approach': 0.1521,
-    'allows': 0.1519, 'stated': 0.1519, 'development': 0.1515, 'commitment': 0.1495,
-    'highlights': 0.1493, 'essential': 0.1483, 'experiences': 0.1480, 'recently': 0.1471,
-    'suggesting': 0.1457, 'market': 0.1447, 'uncertain': 0.1440, 'potentially': 0.1433
-}
-
-OG_WEIGHTS = {
-    'says': 1.0000, 'just': 0.9623, 'people': 0.8774, 'said': 0.8259, 'company': 0.7645,
-    'll': 0.6372, 'make': 0.6237, 'time': 0.5634, 'way': 0.5374, 've': 0.5039,
-    'want': 0.4435, 'like': 0.4426, 'don': 0.4338, 'going': 0.4160, 'really': 0.4126,
-    'use': 0.3769, 'good': 0.3718, 'lot': 0.3710, 'able': 0.3611, 'things': 0.3595,
-    'big': 0.3483, 'doesn': 0.3470, 'right': 0.3453, 'work': 0.3443, 'new': 0.3381,
-    'know': 0.3355, 'think': 0.3218, 'today': 0.3209, 'isn': 0.3039, 'look': 0.3013,
-    'world': 0.2907, 'say': 0.2875, 'best': 0.2825, 'used': 0.2758, 'little': 0.2735,
-    'actually': 0.2724, 'phone': 0.2551, 'thing': 0.2477, 'year': 0.2461, 'come': 0.2328,
-    'told': 0.2315, 'far': 0.2250, 'better': 0.2245, 'didn': 0.2244, 'getting': 0.2209,
-    'help': 0.2193, 'makes': 0.2141, 'got': 0.2139, 'won': 0.2096, 'called': 0.2078,
-    'different': 0.2010, 'verge': 0.2001, 'game': 0.1956, 'looks': 0.1954, 'comes': 0.1953,
-    'years': 0.1935, 'working': 0.1924, 'kind': 0.1899, 'let': 0.1891, 'great': 0.1878,
-    'read': 0.1876, 'number': 0.1868, 'long': 0.1852, 'according': 0.1795, 'coming': 0.1784,
-    'day': 0.1750, 'pretty': 0.1734, 'looking': 0.1685, 'bit': 0.1682, 'place': 0.1677,
-    'start': 0.1667, 'trying': 0.1661, 'sure': 0.1655, 'means': 0.1642, 'course': 0.1641,
-    'week': 0.1637, 'story': 0.1610, 'buy': 0.1589, 'probably': 0.1581, 'play': 0.1561,
-    'using': 0.1554, 'doing': 0.1551, 'hard': 0.1525, 'did': 0.1509, 'money': 0.1497,
-    'point': 0.1472, 'idea': 0.1429, 'end': 0.1425, 'aren': 0.1396, 'fact': 0.1371,
-    'run': 0.1363, 'does': 0.1362, 'case': 0.1331, 'built': 0.1301, 'biggest': 0.1300,
-    'started': 0.1286, 'exactly': 0.1279, 'screen': 0.1277, 'deal': 0.1264, 'apps': 0.1234
-}
+AI_WEIGHTS, OG_WEIGHTS = load_heuristic_weights()
+SIGMOID_K = 0.5
 
 def tokenize(text):
     return re.findall(r'\b[a-z]{2,}\b', text.lower())
@@ -80,7 +38,6 @@ def classify_text_likelihood(text: str) -> float:
     net = ai_score - og_score
     return 1 / (1 + math.exp(-SIGMOID_K * net))
 
-# Wrap words in thick colored underlines based on heuristic
 def highlight_heuristic_words(text: str) -> str:
     parts = re.split(r'(\b[a-z]{2,}\b)', text)
     out = []
@@ -168,7 +125,6 @@ if st.button("Classify", type="primary"):
                 probs = F.softmax(logits, dim=-1).cpu()
                 preds = torch.argmax(probs, dim=-1).cpu()
 
-            # Highlight each sentence and underline heuristic words
             chunks = []
             for i, s in enumerate(sentences):
                 inner = highlight_heuristic_words(s)
@@ -183,7 +139,6 @@ if st.button("Classify", type="primary"):
                 chunks.append(span)
             st.markdown("".join(chunks), unsafe_allow_html=True)
 
-            # Scores
             avg = torch.mean(probs, dim=0)
             model_ai = avg[0].item()
             heuristic_ai = classify_text_likelihood(text)
@@ -191,4 +146,4 @@ if st.button("Classify", type="primary"):
 
             st.subheader(f"🤖 Model AI Likelihood: {model_ai*100:.1f}%")
             st.subheader(f"🛠️ Heuristic AI Likelihood: {heuristic_ai*100:.1f}%")
-            st.subheader(f"⚖️ Combined AI Likelihood: {combined*100:.1f}%")
+            st.subheader(f"⚖️ Combined AI Likelihood: {combined*100:.1f}%")