the difference between of clf.predict(X_train) and clf.labels_

[planetb113]

why:
np.array_equal(clf.predict(X_train), clf.labels_)
and
np.array_equal(clf.decision_function(X_train), clf.decision_scores_)
are both False ?
could any body give me a clue?

[yzhao062]

I am wondering the specific algorithm you are referring to. It is understanble some of them may not show the exact same result. For instance, if contamination is not set to the exact percentage of the outliers in the train set.

[planetb113]

I am wondering the specific algorithm you are referring to. It is understanble some of them may not show the exact same result. For instance, if contamination is not set to the exact percentage of the outliers in the train set.

thanks for your reply, this is my whole script:

from pyod.models.knn import KNN  
import numpy as np

contamination=0.006  
clf = KNN(contamination=contamination)  
clf.fit(X_train)  
x11 = clf.labels_  
x22 = clf.predict(X_train)  
print(np.array_equal(x11, x22))  

is it has something to do with contamination?

[fdewez]

You can execute the following code to get an example of decision_scores_ different from decision_function(X) (unexpected behavior):

import numpy as np
from pyod.models.knn import KNN

rng = np.random.default_rng(42)
X = np.vstack(
   (
      rng.multivariate_normal([-3, -3], [[1, 0], [0, 1]], 300),
      rng.multivariate_normal([3, 2], [[1, 0], [0, 1]], 200)
   )
)

model = KNN()
model.fit(X)
np.testing.assert_array_equal(model.decision_scores_, model.decision_function(X))