from concurrent.futures import ProcessPoolExecutor, as_completed
 
    @staticmethod
    def process_single_video(args):
        """ Helper function for multiprocessing video feature extraction. """
        row, video_base_path, row_index = args
        extractor = VideoFeatureExtractor(use_yolo=False)
        video_path = os.path.join(video_base_path, row["videos"])
        features = extractor.extract_features_from_video(video_path)
        features["time_segment_id"] = row["time_segment_id"]
        return row_index, features
        
                
   def prepare_training_data(self, df: pd.DataFrame, video_base_path: str = "videos") -> pd.DataFrame:
       print("Parallel video feature extractor")
       max_workers = 4
       results = []               
       with ProcessPoolExecutor(max_workers=max_workers) as executor:
          futures = []
          for idx, row in df.iterrows():
              futures.append(executor.submit(self.process_single_video, (row, video_base_path, idx))
              
          for idx, future in enumerate(as_completed(futures)):
              row_index, features = future.result()
              results.append((row_index, features))
              
              if (idx + 1) % 100 == 0:
                  print(f"  Processed video: {idx + 1}/{len(df)}")
     
    results = sorted(results, key=lambda x: x[0])
    video_features_list = [x[1] for x in results]
    # Video özelliklerini birleştir
    ....