def _generate_samples(self, x, epsilon_map):
"""
Generate samples around the current image.
:param x: Sample input with shape as expected by the model.
:type x: `np.ndarray`
:param epsilon_map: Samples drawn from search space
:type epsilon_map: `np.ndarray`
:return: Two arrays of new input samples to approximate gradient
:rtype: `list(np.ndarray)`
"""
minus = clip_and_round(np.repeat(x, self.num_basis, axis=0) - epsilon_map, self.clip_values, self.round_samples)
plus = clip_and_round(np.repeat(x, self.num_basis, axis=0) + epsilon_map, self.clip_values, self.round_samples)
return minus, plus
def predict(self, x: np.ndarray, batch_size: int = 128, **kwargs) -> np.ndarray: # pylint: disable=W0221
"""
Perform prediction of the classifier for input `x`. Rounds results first.
:param x: Features in array of shape (nb_samples, nb_features) or (nb_samples, nb_pixels_1, nb_pixels_2,
nb_channels) or (nb_samples, nb_channels, nb_pixels_1, nb_pixels_2).
:param batch_size: Size of batches.
:return: Array of predictions of shape `(nb_inputs, nb_classes)`.
"""
return self._classifier.predict(clip_and_round(x, self.clip_values, self.round_samples), batch_size=batch_size)
def _generate_samples(self, x: np.ndarray, epsilon_map: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
"""
Generate samples around the current image.
:param x: Sample input with shape as expected by the model.
:param epsilon_map: Samples drawn from search space.
:return: Two arrays of new input samples to approximate gradient.
"""
minus = clip_and_round(
np.repeat(x, self.num_basis, axis=0) - epsilon_map,
self.clip_values,
self.round_samples,
)
plus = clip_and_round(
np.repeat(x, self.num_basis, axis=0) + epsilon_map,
self.clip_values,
self.round_samples,
)
return minus, plus
def _wrap_predict(self, x, batch_size=128):
"""
Perform prediction for a batch of inputs. Rounds results first.
:param x: Test set.
:type x: `np.ndarray`
:param batch_size: Size of batches.
:type batch_size: `int`
:return: Array of predictions of shape `(nb_inputs, nb_classes)`.
:rtype: `np.ndarray`
"""
return self._predict(clip_and_round(x, self.clip_values, self.round_samples), **{"batch_size": batch_size})
def _wrap_predict(self,
x: np.ndarray,
batch_size: int = 128) -> np.ndarray:
"""
Perform prediction for a batch of inputs. Rounds results first.
:param x: Input samples.
:param batch_size: Size of batches.
:return: Array of predictions of shape `(nb_inputs, nb_classes)`.
"""
return self._predict(
clip_and_round(x, self.clip_values, self.round_samples),
**{"batch_size": batch_size})
def _wrap_predict(self, x, logits=False, batch_size=128):
"""
Perform prediction for a batch of inputs. Rounds results first.
:param x: Test set.
:type x: `np.ndarray`
:param logits: `True` if the prediction should be done at the logits layer.
:type logits: `bool`
:param batch_size: Size of batches.
:type batch_size: `int`
:return: Array of predictions of shape `(nb_inputs, self.nb_classes)`.
:rtype: `np.ndarray`
"""
return self._predict(
clip_and_round(x, self.clip_values, self.round_samples), logits,
batch_size)
