V3.0: Add rate-limit resilience (cache+retry) for HF Spaces

app.py

@@ -7,7 +7,7 @@ Risk Engine, Sentiment, Macro, Research Desk, Technical Analysis.
 Bloomberg Terminal aesthetic: black + orange + green + cyan.
 Powered by K2 Think V2 (MBZUAI) for AI analysis.
 """
-import os, json, warnings, math, random
+import os, json, warnings, math, random, time, hashlib, threading
 from datetime import datetime, timedelta
 warnings.filterwarnings('ignore')
 
@@ -56,7 +56,7 @@ class K2ThinkClient:
             return f"🔴 Error: {str(e)[:300]}"
 
 # =============================================================================
-# MARKET DATA
+# MARKET DATA (with caching + retry to handle HF Spaces shared-IP rate limits)
 # =============================================================================
 MARKETS = {
     "US Equities": {"suffix": "", "ex": "AAPL, TSLA, NVDA, SPY, QQQ"},
@@ -72,15 +72,41 @@ MARKETS = {
     "Indices": {"suffix": "", "ex": "^GSPC, ^DJI, ^IXIC, ^FTSE"},
 }
 
+# In-memory cache with TTL for yfinance data (mitigates HF Spaces shared-IP rate limits)
+_FETCH_CACHE = {}
+_FETCH_LOCK = threading.Lock()
+
+def _cache_key(ticker, period, interval):
+    import hashlib
+    return hashlib.md5(f"{ticker.upper().strip()}|{period}|{interval}".encode()).hexdigest()
+
 def fetch(ticker, period="1y", interval="1d"):
-    try:
-        stock = yf.Ticker(ticker.upper().strip())
-        df = stock.history(period=period, interval=interval)
-        if df.empty:
-            return None, f"No data for '{ticker}'."
-        return df, stock.info
-    except Exception as e:
-        return None, f"Error: {str(e)[:200]}"
+    key = _cache_key(ticker, period, interval)
+    with _FETCH_LOCK:
+        if key in _FETCH_CACHE:
+            entry = _FETCH_CACHE[key]
+            if time.time() - entry['ts'] < 60:
+                return entry['data'], entry['info']
+    
+    t = ticker.upper().strip()
+    last_err = ""
+    for attempt in range(3):
+        try:
+            time.sleep(attempt * 1.5)
+            stock = yf.Ticker(t)
+            df = stock.history(period=period, interval=interval, auto_adjust=False)
+            if df.empty:
+                return None, f"No data for '{ticker}'."
+            info = stock.info if hasattr(stock, 'info') else {}
+            with _FETCH_LOCK:
+                _FETCH_CACHE[key] = {'ts': time.time(), 'data': df.copy(), 'info': info}
+            return df, info
+        except Exception as e:
+            last_err = str(e)
+            if 'Too Many Requests' in last_err or 'Rate limited' in last_err:
+                continue
+            break
+    return None, f"Error fetching '{ticker}': {last_err[:200]}. Yahoo Finance rate-limits shared IPs (HF Spaces). Try again in 30s."
 
 # =============================================================================
 # TECHNICAL INDICATORS
@@ -538,7 +564,8 @@ def crypto_arbitrage(coins):
         if not coin: continue
         sym = f"{coin}-USD"
         try:
-            df = yf.Ticker(sym).history(period="1d", interval="1m")
+            time.sleep(0.5)
+            df = yf.Ticker(sym).history(period="1d", interval="1m", auto_adjust=False)
             if not df.empty:
                 results.append({
                     'Coin': coin,
@@ -549,11 +576,12 @@ def crypto_arbitrage(coins):
                     'Volume': f"{df['Volume'].sum():,.0f}",
                     'Spread %': f"{((df['High'].iloc[-1]/df['Low'].iloc[-1]-1)*100):.3f}%"
                 })
-        except:
-            pass
+        except Exception as e:
+            if 'Rate' not in str(e) and 'Too Many' not in str(e):
+                pass  # ignore other errors
     
     if not results:
-        return None, "Could not fetch crypto data. Try BTC, ETH, SOL."
+        return None, "Could not fetch crypto data. Yahoo Finance may be rate-limiting. Try BTC, ETH, SOL, or wait 30s."
     
     df = pd.DataFrame(results)
     # Arbitrage heatmap (simulated cross-exchange spreads)
@@ -736,15 +764,13 @@ def sentiment_analyzer(ticker):
 def macro_analysis():
     macros = {}
     for t, name in [('^GSPC','S&P 500'),('^IXIC','Nasdaq'),('^TNX','10Y Treasury'),('GC=F','Gold'),('CL=F','Oil'),('EURUSD=X','EUR/USD'),('DX-Y.NYB','DXY Dollar'),('BTC-USD','Bitcoin')]:
-        try:
-            df = yf.Ticker(t).history(period='3mo')
-            if not df.empty:
-                macros[name] = {'price': df['Close'].iloc[-1], '1m': (df['Close'].iloc[-1]/df['Close'].iloc[0]-1)*100,
-                                '3m': (df['Close'].iloc[-1]/df['Close'].iloc[max(0,len(df)-63)]-1)*100 if len(df)>63 else 0}
-        except: pass
+        df, info, err = fetch(t, "3mo")
+        if df is not None and not df.empty:
+            macros[name] = {'price': df['Close'].iloc[-1], '1m': (df['Close'].iloc[-1]/df['Close'].iloc[0]-1)*100,
+                            '3m': (df['Close'].iloc[-1]/df['Close'].iloc[max(0,len(df)-63)]-1)*100 if len(df)>63 else 0}
     
     if not macros:
-        return "Could not fetch macro data."
+        return None, "Could not fetch macro data."
     
     fig = go.Figure()
     names = list(macros.keys())
@@ -1113,7 +1139,7 @@ def build_app():
             ### 6. RISK ENGINE
             | Feature | Jane Street Practice |
             |---------|---------------------|
-            | Parametric + Historical VaR | Dual methodology for compliance |
+            | Parametric + Historical VaR | Dual methodology for regulatory compliance |
             | Stress testing | 2008, COVID-2020, 2022 rate hike scenarios |
             | Correlation breakdown | Crisis: correlations -> 1 |
             | Student-t tails | Fat-tail distribution modeling |
@@ -1121,17 +1147,17 @@ def build_app():
             ### 7. SENTIMENT ANALYZER
             | Feature | Jane Street Practice |
             |---------|---------------------|
-            | Multi-source NLP | Bloomberg, SEC filings, Twitter, Reddit |
-            | NER | Company/executive/product mention extraction |
-            | Temporal momentum | Improving vs deteriorating sentiment |
+            | Multi-source NLP pipeline | Bloomberg headlines, SEC filings, Twitter, Reddit |
+            | Named Entity Recognition | Company/executive/product mention extraction |
+            | Temporal analysis | Improving vs deteriorating sentiment |
             | Alpha factor | Sentiment surprise IC = 0.3-0.5 |
             
             ### 8. MACRO DASHBOARD
             | Feature | Jane Street Practice |
             |---------|---------------------|
             | Growth/Inflation quadrant | Bridgewater All Weather framework |
-            | Dollar regime | DXY >100 = risk-off, EM stress |
-            | Yield curve | 10Y-2Y inversion = recession (9/10) |
+            | Dollar regime | DXY > 100 = risk-off, EM stress |
+            | Yield curve | 10Y-2Y inversion = recession (9/10 accuracy) |
             | Cross-asset momentum | Asness value/momentum factors |
             
             ### Stack