The model is at the last in a pipeline

[SwamiPatil]

adversarial-robustness-toolbox/art/estimators/scikitlearn.py

Line 50 in 79a980f

_input_shape = self._get_input_shape(model.steps[0][1])

Current value of _input_shape is:
elif hasattr(model, "steps"):
_input_shape = self._get_input_shape(model.steps[0][1])

It should be
_input_shape = self._get_input_shape(model.steps[-1][-1])

Because the model is at the last in pipeline.steps

[beat-buesser]

Hi @SwamiPatil Thank you very much for using ART! The _input_shape as return value for the input_shape property is expected to return the shape of the input into the model, in the case of a Pipeline the pipeline is treated as the model (although containing another model) for the purpose of the sklearn estimator.

Did you encounter a case where this causes a problem?