DIF model: duplicate normalization

[Valentina]

On the DIF (Deep Isolation Forest) model, in the fit function, the variable X is normalized. Then it is passed to decision_function to compute the decision_scores_, where it is normalized again. This results in a mismatch between decision_scores_ and scores obtained by calling decision_function(X) on the same X.

Normalization:

pyod/pyod/models/dif.py

Lines 173 to 175 in 690a0f2

 self.minmax_scaler = MinMaxScaler() 
 self.minmax_scaler.fit(X) 
 X = self.minmax_scaler.transform(X) 

decision_function call:

pyod/pyod/models/dif.py

Lines 215 to 216 in 690a0f2

 self.decision_scores_ = self.decision_function(X) 
 self._process_decision_scores() 

Normalization in decision_function:

pyod/pyod/models/dif.py

Line 241 in 690a0f2

 X = self.minmax_scaler.transform(X)