v.1.21

app.py

@@ -80,6 +80,29 @@ class EventDetector:
             logger.error(f"Error in EventDetector initialization: {e}")
             raise
 
+    def get_sentiment_label(self, result):
+        """
+        Convert model output to standardized sentiment label
+        """
+        try:
+            # Handle different model output formats
+            if isinstance(result, dict):
+                label = result.get('label', '').lower()
+            else:
+                return "Neutral"
+
+            # Map different model outputs to standard labels
+            if label in ['positive', 'pos', 'positive tone']:
+                return "Positive"
+            elif label in ['negative', 'neg', 'negative tone']:
+                return "Negative"
+            else:
+                return "Neutral"
+
+        except Exception as e:
+            logger.error(f"Error in get_sentiment_label: {e}")
+            return "Neutral"
+
     @spaces.GPU(duration=30)
     def initialize_models(self):
         if self.initialized:
@@ -144,17 +167,56 @@ class EventDetector:
             self.cleanup()
             raise
 
-    def cleanup(self):
-        """Clean up GPU resources"""
+    @spaces.GPU(duration=20)
+    def analyze_sentiment(self, text):
         try:
-            self.model = None
-            self.finbert = None
-            self.roberta = None
-            self.finbert_tone = None
-            torch.cuda.empty_cache()
-            self.initialized = False
+            if not self.initialized:
+                if not self.initialize_models():
+                    return "Neutral"
+            
+            current_time = time.time()
+            if current_time - self.last_gpu_use < 2:
+                time.sleep(2)
+            
+            truncated_text = text[:500]
+            results = []
+            
+            try:
+                inputs = [truncated_text]
+                sentiment_results = []
+                
+                # Process each model separately with delay
+                if self.finbert:
+                    finbert_result = self.finbert(inputs, truncation=True, max_length=512)[0]
+                    results.append(self.get_sentiment_label(finbert_result))
+                    time.sleep(0.5)
+                
+                if self.roberta:
+                    roberta_result = self.roberta(inputs, truncation=True, max_length=512)[0]
+                    results.append(self.get_sentiment_label(roberta_result))
+                    time.sleep(0.5)
+                
+                if self.finbert_tone:
+                    finbert_tone_result = self.finbert_tone(inputs, truncation=True, max_length=512)[0]
+                    results.append(self.get_sentiment_label(finbert_tone_result))
+                
+                # Get majority vote
+                if results:
+                    sentiment_counts = pd.Series(results).value_counts()
+                    final_sentiment = sentiment_counts.index[0] if sentiment_counts.iloc[0] >= 2 else "Neutral"
+                else:
+                    final_sentiment = "Neutral"
+                
+                self.last_gpu_use = time.time()
+                return final_sentiment
+                
+            except Exception as e:
+                logger.error(f"Model inference error: {e}")
+                return "Neutral"
+            
         except Exception as e:
-            logger.error(f"Error in cleanup: {e}")
+            logger.error(f"Sentiment analysis error: {e}")
+            return "Neutral"
 
     @spaces.GPU(duration=20)
     def detect_events(self, text, entity):
@@ -211,83 +273,17 @@ class EventDetector:
             logger.error(f"Event detection error: {str(e)}")
             return "Нет", f"Error: {str(e)}"
 
-    @spaces.GPU(duration=20)
-    def analyze_sentiment(self, text):
+    def cleanup(self):
+        """Clean up GPU resources"""
         try:
-            if not self.initialized:
-                if not self.initialize_models():
-                    return "Neutral"
-            
-            current_time = time.time()
-            if current_time - self.last_gpu_use < 2:
-                time.sleep(2)
-            
-            truncated_text = text[:500]
-            results = []
-            
-            try:
-                inputs = [truncated_text]
-                sentiment_results = []
-                
-                # Process each model separately with delay
-                if self.finbert:
-                    finbert_result = self.finbert(inputs, truncation=True, max_length=512)[0]
-                    results.append(self.get_sentiment_label(finbert_result))
-                    time.sleep(0.5)
-                
-                if self.roberta:
-                    roberta_result = self.roberta(inputs, truncation=True, max_length=512)[0]
-                    results.append(self.get_sentiment_label(roberta_result))
-                    time.sleep(0.5)
-                
-                if self.finbert_tone:
-                    finbert_tone_result = self.finbert_tone(inputs, truncation=True, max_length=512)[0]
-                    results.append(self.get_sentiment_label(finbert_tone_result))
-                
-                # Get majority vote
-                if results:
-                    sentiment_counts = pd.Series(results).value_counts()
-                    final_sentiment = sentiment_counts.index[0] if sentiment_counts.iloc[0] >= 2 else "Neutral"
-                else:
-                    final_sentiment = "Neutral"
-                
-                self.last_gpu_use = time.time()
-                return final_sentiment
-                
-            except Exception as e:
-                logger.error(f"Model inference error: {e}")
-                return "Neutral"
-            
+            self.model = None
+            self.finbert = None
+            self.roberta = None
+            self.finbert_tone = None
+            torch.cuda.empty_cache()
+            self.initialized = False
         except Exception as e:
-            logger.error(f"Sentiment analysis error: {e}")
-            return "Neutral"
-
-def create_visualizations(df):
-    if df is None or df.empty:
-        return None, None
-        
-    try:
-        sentiments = df['Sentiment'].value_counts()
-        fig_sentiment = go.Figure(data=[go.Pie(
-            labels=sentiments.index,
-            values=sentiments.values,
-            marker_colors=['#FF6B6B', '#4ECDC4', '#95A5A6']
-        )])
-        fig_sentiment.update_layout(title="Распределение тональности")
-        
-        events = df['Event_Type'].value_counts()
-        fig_events = go.Figure(data=[go.Bar(
-            x=events.index,
-            y=events.values,
-            marker_color='#2196F3'
-        )])
-        fig_events.update_layout(title="Распределение событий")
-        
-        return fig_sentiment, fig_events
-        
-    except Exception as e:
-        logger.error(f"Visualization error: {e}")
-        return None, None
+            logger.error(f"Error in cleanup: {e}")
 
 @spaces.GPU
 def process_file(file_obj):
@@ -399,7 +395,7 @@ def create_interface():
     control = ProcessControl()
     
     with gr.Blocks(theme=gr.themes.Soft()) as app:
-        gr.Markdown("# AI-анализ мониторинга новостей v.1.20")
+        gr.Markdown("# AI-анализ мониторинга новостей v.1.21")
         
         with gr.Row():
             file_input = gr.File(