normalization of drift

drift measure is now between 0. and 1. (%)
AxeldeRomblay · Aug 2, 2017 · 682f09c · 682f09c
1 parent b1bbb8e
commit 682f09c
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 8 deletions.
diff --git a/python-package/mlbox/preprocessing/drift/drift_estimator.py b/python-package/mlbox/preprocessing/drift/drift_estimator.py
@@ -157,7 +157,7 @@ def score(self):
 
         """Returns the global drift measure between two datasets.
 
-         0.50 = No drift. 1.00 = Maximal Drift
+         0. = No drift. 1. = Maximal Drift
 
         Returns
         -------
@@ -177,8 +177,7 @@ def score(self):
                 S.append(roc_auc_score(self.__cible.iloc[test_index],
                                        self.__pred[test_index]))
 
-            # TODO: return the mean of mirrored AUC metrics
-            return max(np.mean(S), 1-np.mean(S))
+            return (max(np.mean(S), 1-np.mean(S))-0.5) * 2
 
         else:
             raise ValueError('Call the fit function before !')

diff --git a/python-package/mlbox/preprocessing/drift/drift_threshold.py b/python-package/mlbox/preprocessing/drift/drift_threshold.py
@@ -59,7 +59,7 @@ class DriftThreshold():
     ----------
     threshold : float, defaut = 0.6
         The drift threshold (univariate drift below are kept)
-        Must be between 0.5 and 1.
+        Must be between 0. and 1.
 
     subsample : float, defaut = 1.
         Subsampling parameter for the datasets.

diff --git a/python-package/mlbox/preprocessing/drift_thresholder.py b/python-package/mlbox/preprocessing/drift_thresholder.py
@@ -20,9 +20,10 @@ class Drift_thresholder():
 
     Parameters
     ----------
-    threshold : float, defaut = 0.9
-        Threshold used to deletes variables and ids. Must be between 0.5 and 1.
-        The lower the more you keep non-drifting/stable variables.
+    threshold : float, defaut = 0.6
+        Drift threshold under which features are kept. Must be between 0. and 1.
+        The lower the more you keep non-drifting/stable variables: a feature with
+        a drift measure of 0. is very stable and a one with 1. is highly unstable.
 
     inplace : bool, default = False
         If True, train and test datasets are transformed. Returns self.
@@ -37,7 +38,7 @@ class Drift_thresholder():
     """
 
     def __init__(self,
-                 threshold=0.8,
+                 threshold=0.6,
                  inplace=False,
                  verbose=True,
                  to_path="save"):