bumped toolbox

algo/curve_anomaly/cont_det/online_detector.py

@@ -60,9 +60,8 @@ def check(self, value, timestamp):
             util.logger.debug("Model is not available. Skip inference")
             return False, ''
 
-        util.logger.debug(f"First Data Point: {self.data_list[0]} - Last Data Point: {self.data_list[-1]}")
-        if self.data_list[-1][0] - self.data_list[0][0] < pd.Timedelta(4, "h"):
-            util.logger.debug("Not enough data for inference. Need at least 4 hours")
+
+        if not self.inference_shall_start():
             return False, ''
 
         anomalies, self.reconstruction_errors = self.test(self.data_list, self.model)
@@ -78,7 +77,14 @@ def check(self, value, timestamp):
             return True, self.create_result(f'In der Zeit seit {str(time_window_start)} wurde eine Anomalie im Lastprofil festgestellt.', str(time_window_start), "continous_device")
 
         return False, ''
-
+
+    def inference_shall_start(self):
+        util.logger.debug(f"First Data Point: {self.data_list[0]} - Last Data Point: {self.data_list[-1]}")
+        if self.data_list[-1][0] - self.data_list[0][0] < pd.Timedelta(4, "h"):
+            util.logger.debug("Not enough data for inference. Need at least 4 hours")
+            return False
+        return True
+
     def start_training(self, timestamp):
         topic_name, path_to_time, path_to_value = self._get_input_topic()
         job_request = {
@@ -88,7 +94,7 @@ def start_training(self, timestamp):
                     "window_length": 205,
                     "batch_size": 1,
                     "lr": 0.0001,
-                    "num_epochs": 20,
+                    "num_epochs": 2, # TODO increase prod
                     "loss": "MSE",
                     "op": "Adam",
                     "latent_dims": 32,
@@ -121,7 +127,6 @@ def start_training(self, timestamp):
         }
         self.trainer.start_training(job_request)
         self.last_training_time = timestamp
-
 
     def load_model(self):
         # TODO christoph doch lieber model abspeichern beim stoppen und laden 

algo/operator.py

@@ -35,19 +35,21 @@ def parse_bool(value):
 
 class CustomConfig(Config):
     data_path = "/opt/data"
-    check_data_anomalies: bool = False
-    check_data_extreme_outlier: bool = True
-    check_data_schema: bool = True
-    check_receive_time_outlier: bool = True
-    check_consumption: bool = False
-    init_phase_length: float = 2
+    check_data_anomalies: bool = False # Detect complex anomalies based on the curve are detected
+    check_data_extreme_outlier: bool = True # Detect extreme point outlier based on simple statistics
+    check_data_schema: bool = True # Detect wrong input
+    check_receive_time_outlier: bool = True # Detect anomalies in the time between inputs
+    check_consumption: bool = False #
+    init_phase_length: float = 2 # How long the complete operator is init phase
     init_phase_level: str = "d"
-    train_interval: float = 14
+    train_interval: float = 14 # How long curve explorere waits until it starts training & optional retraining
     train_level: str = "d"
-    retrain: bool = False
-    ml_trainer_url: str = "http://ml-trainer-svc.trainer:5000"
-    mlflow_url: str = "http://mlflow-svc.mlflow:5000"
-    curve_detector_training_mode: str = "offline"
+    inference_time: float = 1 # How long operator waits until it starts inference. Should match with the model architecture
+    inference_time_level: str = "s"
+    retrain: bool = False # Only train once in the beginning or retrain periodically
+    ml_trainer_url: str = "http://ml-trainer-svc.trainer:5000" # URL to the Trainer API
+    mlflow_url: str = "http://mlflow-svc.mlflow:5000" # URL for MLFlow to download the model
+    curve_detector_training_mode: str = "offline" # Train inside the operator or use the training platform
 
     def __init__(self, d, **kwargs):
         super().__init__(d, **kwargs)

requirements.txt

@@ -1,7 +1,7 @@
 git+https://github.com/SENERGY-Platform/[email protected]
-toolbox[anomaly_detection] @ git+https://github.com/SENERGY-Platform/[email protected].74
+toolbox[anomaly_detection] @ git+https://github.com/SENERGY-Platform/[email protected].76
 
-pandas<2
+pandas==2.2.2
 kazoo<3
 astral
 pytz