how to attack a custom model that isn't a neural network

[teenaxta]

I am trying to see how attacks work or how noise is added in various attacks. For that I have a made a simple class:

`class quadrant():
def predict(x):
if x < 0:
return 0
else:
return 1

new_model = quadrant()`

Issue is that every attack asks for an estimator built using one of ARTs estimators. I dont want to give it a neural network. I just want to see how a 1D point oscillates

[beat-buesser]

Hi @teenaxta For such applications you could use ART's BlackBoxClassifier estimator and provide it your function predict as input.