Update app.py

app.py

@@ -1,40 +1,109 @@
+# app_news_sentiment_firebase.py
 import feedparser
 import pandas as pd
 import gradio as gr
 from transformers import pipeline
+from datetime import datetime
+import firebase_admin
+from firebase_admin import credentials, db
+import os
+import json
 
-sentiment = pipeline("text-classification", model="cardiffnlp/twitter-roberta-base-sentiment-latest")
+# --- Firebase Configuration ---
+# IMPORTANT: Set these environment variables in your deployment environment (e.g., Hugging Face Spaces Secrets)
+# 1. FIRESTORE_SA_KEY: The full JSON content of your Firebase service account key.
+# 2. FIREBASE_DB_URL: The URL to your Firebase Realtime Database.
+SA_KEY_JSON = os.environ.get('FIRESTORE_SA_KEY')
+DB_URL = os.environ.get('FIREBASE_DB_URL')
 
+class NewsRTDBLogger:
+    def __init__(self, db_ref_name='news_sentiment'):
+        self.ref = None
+        if not all([SA_KEY_JSON, DB_URL]):
+            print("NEWS LOGGER: Firebase secrets not set. Logger is disabled.")
+            return
+        try:
+            # The service account key is loaded from a JSON string in the environment variable
+            cred_dict = json.loads(SA_KEY_JSON)
+            cred = credentials.Certificate(cred_dict)
+            if not firebase_admin._apps:
+                firebase_admin.initialize_app(cred, {'databaseURL': DB_URL})
+            self.ref = db.reference(db_ref_name)
+            print(f"NEWS LOGGER: Successfully connected to Firebase RTDB at '{db_ref_name}'.")
+        except Exception as e:
+            print(f"NEWS LOGGER: CRITICAL ERROR - Failed to initialize: {e}")
+
+    def log_sentiment(self, pub_date, headline, sentiment, confidence):
+        if not self.ref:
+            return
+        try:
+            # Create a unique key based on the timestamp to keep logs ordered
+            log_entry = {
+                "timestamp_published": pub_date,
+                "headline": headline,
+                "sentiment": sentiment,
+                "confidence_score": float(confidence)
+            }
+            self.ref.push(log_entry)
+            # print(f"NEWS LOGGER: Logged '{headline}'")
+        except Exception as e:
+            print(f"NEWS LOGGER: CRITICAL ERROR - Could not write sentiment log: {e}")
+
+# --- Initialize Global Objects ---
+sentiment_pipeline = pipeline("text-classification", model="cardiffnlp/twitter-roberta-base-sentiment-latest")
+news_logger = NewsRTDBLogger() # Initialize the logger
+
+# --- Core Functions ---
 def fetch_news():
+    """Fetches the latest news headlines from the FXStreet RSS feed."""
     url = "https://www.fxstreet.com/rss/news"
     feed = feedparser.parse(url)
-    headlines = feed.entries[:7]
+    # Fetch a few more headlines to ensure we get recent, unique ones
+    headlines = feed.entries[:15]
     return headlines
 
 def analyze_news():
+    """
+    Analyzes news sentiment, logs it to Firebase, and returns a DataFrame for display.
+    """
     headlines = fetch_news()
     results = []
+    
     for h in headlines:
         text = h.title
-        pred = sentiment(text)[0]
+        # Get the publication date, fallback to now if not present
+        pub_date = h.get("published", datetime.utcnow().isoformat())
+        
+        # Perform sentiment analysis
+        pred = sentiment_pipeline(text)[0]
         label = pred["label"].lower()
         score = round(pred["score"], 3)
+        
         if label == "positive":
-            m = "Bullish"
+            market_sentiment = "Bullish"
         elif label == "negative":
-            m = "Bearish"
+            market_sentiment = "Bearish"
         else:
-            m = "Neutral"
-        results.append([text, label, score, m])
-    df = pd.DataFrame(results, columns=["Headline", "Raw Label", "Confidence", "Market Sentiment"])
-    return df
+            market_sentiment = "Neutral"
+        
+        # Log the result to Firebase
+        news_logger.log_sentiment(pub_date, text, market_sentiment, score)
+        
+        results.append([pub_date, text, market_sentiment, score])
+        
+    # Create DataFrame for Gradio output, showing the 7 most recent
+    df = pd.DataFrame(results, columns=["Date", "Headline", "Market Sentiment", "Confidence"])
+    return df.head(7)
 
+# --- Gradio Interface ---
 demo = gr.Interface(
     fn=analyze_news,
     inputs=None,
-    outputs=gr.Dataframe(),
-    title="📊 Forex Sentiment Agent",
-    description="Live Forex headlines sentiment (Bullish / Bearish / Neutral)."
+    outputs=gr.Dataframe(headers=["Date", "Headline", "Market Sentiment", "Confidence"]),
+    title="📊 Forex Sentiment Agent (Firebase Edition)",
+    description="Live Forex headlines sentiment (Bullish / Bearish / Neutral). Now logging results directly to Firebase.",
+    allow_flagging='never'
 )
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()