add EntireAnomalyDetector, LatestAnomalyDetector

anomaly-detector/anomaly_detector/__init__.py

@@ -1,2 +1,2 @@
 from .multivariate.model import MultivariateAnomalyDetector
-from .univariate.univariate_anomaly_detection import UnivariateAnomalyDetector
+from .univariate.univariate_anomaly_detection import UnivariateAnomalyDetector, EntireAnomalyDetector, LatestAnomalyDetector

anomaly-detector/anomaly_detector/univariate/model/detect_model.py

@@ -92,7 +92,7 @@ def merge_with_delta(results, func, args):
 
 class AnomalyDetectionModel:
     def __init__(self, series=None, max_anomaly_ratio=DEFAULT_MAX_RATIO, alpha=DEFAULT_ALPHA, threshold=DEFAULT_THRESHOLD, granularity=DEFAULT_GRANULARITY_NONE, interval=None, indices=None,
-                 fill_up_mode=DEFAULT_FILL_UP_MODE, fixed_value_to_fill=None, need_trend=False, need_spectrum_period=False, detector=dict(), detect_mode=None):
+                 fill_up_mode=DEFAULT_FILL_UP_MODE, fixed_value_to_fill=None, need_trend=False, need_spectrum_period=False, detector=dict()):
         self.__values = [float(item["value"]) for item in series]
         self.__max_ratio = max_anomaly_ratio
         self.__alpha = alpha

anomaly-detector/anomaly_detector/univariate/univariate_anomaly_detection.py

@@ -14,11 +14,12 @@
 from anomaly_detector.univariate.util.refine import get_margins
 
 class UnivariateAnomalyDetector(BaseAnomalyDetector):
-    def __init__(self):
+    def __init__(self, detect_mode = None):
 
         super(UnivariateAnomalyDetector, self).__init__()
-        self.error_msg = None
-        self.error_code = 'BadArgument'
+        self._error_msg = None
+        self._error_code = 'BadArgument'
+        self._detect_mode = detect_mode
 
     def is_timestamp_ascending(self, series):
         count = len(series)
@@ -236,22 +237,24 @@ def parse_arg(self, data, params):
     def predict(self, context, data: pd.DataFrame, params: Optional[Dict[str, Any]] = None):
 
         data, model_params = self.parse_arg(data, params)
+        if self._detect_mode is None:
+            self._detect_mode = params['detect_mode']
 
-        if self.error_msg is not None:
-            raise AnomalyDetectionRequestError(error_code=self.error_code, error_msg=self.error_msg)
+        if self._error_msg is not None:
+            raise AnomalyDetectionRequestError(error_code=self._error_code, error_msg=self.error_msg)
 
         detector = AnomalyDetectionModel(series=data, **model_params)
 
         results, period, spectrum_period, model_id, do_fill_up = detector.detect(
-            period=params.get('period',DEFAULT_PERIOD), last_value=data[-1] if params['detect_mode'] == DetectType.LATEST else None)
+            period=params.get('period',DEFAULT_PERIOD), last_value=data[-1] if self._detect_mode == DetectType.LATEST else None)
 
         try:
             results["timestamp"] = [x['timestamp'] for x in data]
         except KeyError:
             results["timestamp"] = [x for x in range(len(data))]
 
         results_items= {}
-        if params['detect_mode'] == DetectType.ENTIRE:
+        if self._detect_mode == DetectType.ENTIRE:
 
             results = results.sort_index()        
             expected_value, upper_margin, lower_margin, anomaly_neg, anomaly_pos, anomaly, severity, boundary_units, anomaly_scores \
@@ -305,4 +308,13 @@ def predict(self, context, data: pd.DataFrame, params: Optional[Dict[str, Any]]
                     SuggestedWindow: suggested_window
                 }}             
             ]
-        return result_list
+        return result_list
+
+
+class EntireAnomalyDetector(UnivariateAnomalyDetector):
+    def __init__(self):
+        super(EntireAnomalyDetector, self).__init__(DetectType.ENTIRE)
+
+class LatestAnomalyDetector(UnivariateAnomalyDetector):
+    def __init__(self):
+        super(LatestAnomalyDetector, self).__init__(DetectType.LATEST)

anomaly-detector/tests/e2e_test_2.py

@@ -1,7 +1,7 @@
 import mlflow
 import numpy as np
 import pandas as pd
-from anomaly_detector import UnivariateAnomalyDetector
+from anomaly_detector import EntireAnomalyDetector
 from mlflow.models import infer_signature
 
 mlflow.set_tracking_uri(uri="http://127.0.0.1:5000")
@@ -11,7 +11,6 @@ def main():
     mlflow.set_experiment("MLflow Quickstart 2")
 
     params = {
-        "detect_mode": "entire",
         "granularity": "monthly", 
         "maxAnomalyRatio": 0.25, 
         "sensitivity": 95, 
@@ -22,7 +21,7 @@ def main():
         mlflow.log_params(params)
         mlflow.set_tag("Training Info", "Univariate Anomaly Detector")
 
-        model = UnivariateAnomalyDetector()
+        model = EntireAnomalyDetector()
 
         signature = infer_signature(params=params)
         print(model)