g-safer 0.2

safer package

SAFER

This guide provides SAFER model module

Baseline

• Baseline
  • data_processing_m1
    • crf_data.py
    • location_data.py
    • sensor_data.py
  • data_processing_m2
    • crf_data.ppy
    • location_data.py
    • sensor_data.py
  • model1
    • dataloader.py
    • model.py
    • predictor.py
  • model2
    • dataloader.py
    • model.py
    • preictor.py
  • setup.py
  • README.md

How To Use

Start pip install

• data processing

  1. location

     You can load location data from preprocess it as follows:

  
    from location_processor import LocationProcessor
  
    file_path = ''
    processed_location_data = LocationProcessor.load_data_from_csv(file_path)
  
    location_dict = {
        (37.7749, -122.4194): 'ward',
        (34.0522, -118.2437): 'hallway',
        (40.7128, -74.0060): 'other',
    }
  
    labeled_location_data = LocationProcessor.assign_location_labels(processed_data, location_dict)
  
  
    

  2. sensor

     you can load sensor data from preprocess it as follows :

     
      from sensor_processor import SensorDataProcessor
     
      file_path = ''
      sensing_data = SensorDataProcessor.load_sensing_data(file_path)
      sensing_data = SensorDataProcessor.process_sensing_data(sensing_data)
      sensing_data = SensorDataProcessor.aggregate_sensing_data(sensing_data)
      sensing_data = SensorDataProcessor.reorganize_column_names(sensing_data)
      

  3. patient data (2 type of data)

     you can load patient data from preprocess it as follows :

     
      from crf_data import DataProcessor
      status_file_path = ''
      trait_fiile_path = ''
         
      processor = DataProcessor()
     
      processor.load_data(
          location_file=labeled_location_data,
          sensor_file=sensing_data,
          crf_file= status_file_path ,
          trait_file= trait_fiile_path
      )
     
      processor.merge_location_and_sensor()
      processor.process_crf_data()
      processor.merge_trait_data()
     
     
     
      suicide_flags = [
          ('patient_key', pd.to_datetime('2023-12-02 00:00:00')),
          .
          .
      ]
     
     
     
      final_data = processor.clean_and_set_suicide_flag(suicide_flags)
      final_data = filter_data_for_self_harm_and_random(final_data, suicide_flags)
     
      

• model

  you can predict m1 from preprocess it as follows :

  • m1

  
        from model1.model import TemporalFusionTransformer
        from model1.predictor import PredictionHandler
  
        data_paths = ['']
        predictor = PredictionHandler(data_paths, batch_size=16, device='cpu')
        predictions = predictor.predict()
  
     

  • m2

  
        from model2.predictor import Predictor
        import torch
        from model2.model import CNNGRUClassificationModel
  
        data_path = ''  
  
        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
  
        predictor = Predictor(device=device)
  
        data_loader = predictor.preprocess_data(data_path)
  
        predictions = predictor.predict(data_loader)
  
        print(predictions)