063c572b5b09fb8492beee938589cf5984f84926,art/estimators/classification/ensemble.py,EnsembleClassifier,class_gradient,#EnsembleClassifier#Any#Any#Any#,234

Before Change

                 `(batch_size, 1, input_shape)` when `label` parameter is specified. If `raw=True`, an additional
                 dimension is added at the beginning of the array, indexing the different classifiers.
        
        grads = np.array(
            [
                self._classifier_weights[i] * self._classifiers[i].class_gradient(x, label)
                for i in range(self._nb_classifiers)
            ]
        )
        if raw:
            return grads

        return np.sum(grads, axis=0)

    def loss_gradient(self, x: np.ndarray, y: np.ndarray, raw: bool = False, **kwargs) -> np.ndarray:

After Change

                    % (str(classifier.input_shape), str(classifiers[0].input_shape))
                )

        self._input_shape = classifiers[0].input_shape
        self._nb_classes = classifiers[0].nb_classes

        // Set weights for classifiers
        if classifier_weights is None:
            classifier_weights = np.ones(self._nb_classifiers) / self._nb_classifiers
        self._classifier_weights = classifier_weights

        // check for consistent channels_first in ensemble members
        for i_cls, cls in enumerate(classifiers):
            if cls.channels_first != self.channels_first:
                raise ValueError(
                    "The channels_first boolean of classifier {} is {} while this ensemble expects a "
                    "channels_first boolean of {}. The channels_first booleans of all classifiers and the "

Instances