minot documentation updates based on the MR comments

Author: sayanchk

luminaire/exploration/data_exploration.py

@@ -810,6 +810,7 @@ def kf_naive_outlier_detection(self, input_series, idx_position):
 
         residuals = truncated_series - filtered_state_means[:, 0]
 
+        # Catching marginal anomalies to avoid during training
         is_anomaly = residuals[idx_position] < np.mean(residuals) \
                      - (1 * np.sqrt(filtered_state_covariance)[idx_position][0][0]) \
                      or residuals[idx_position] > np.mean(residuals) \

luminaire/model/lad_structural.py

@@ -443,7 +443,8 @@ def _validate_model(self, data, hyper_params, result):
         :param pandas.DataFrame data: Input time series data
         :param dict hyper_params: Hyperparameters for model training
         :param dict result: Training outputs
-        :return: A validation flag for model to publish
+        :return: A validation flag for model to publish. The function returns True when the validation is successful,
+        False otherwise.
 
         :rtype: bool
         """
@@ -472,6 +473,9 @@ def _validate_model(self, data, hyper_params, result):
             if result['Success']:
                 predictions.append(result['Prediction'])
 
+        # We perform Levene test, Mann-Whitney U test, and Grubb test between the predictions and the baseline
+        # to understand if the are widely different. The two arrays need to be of the same length in order to
+        # perform these tests.
         if len(predictions) == len(baseline):
             N = len(predictions)
 
@@ -547,6 +551,7 @@ def train(self, data, optimize=False, validate=False, **kwargs):
 
         success = False if 'ErrorMessage' in result else True
 
+        # Model validation step to check underfit due to misspecification of hyperparameters to some other reasons.
         if success and validate and not optimize:
             passed = self._validate_model(data=data, hyper_params=self.hyper_params, result=result)
             if not passed:

requirements.txt

@@ -1,4 +1,6 @@
+# installing latest changes for bayesian-changepoint-detection since the scipy support is not available in pypi yet
 git+https://github.com/hildensia/bayesian_changepoint_detection@2dd95f5c1d028116899a842ccb3baa173f9d5be9
+
 changepy>=0.3.1
 hyperopt>=0.1.2
 numpy>=1.17.5