def gen_imbalanced_data()

in datasets/ImbalanceCIFAR.py [0:0]

• 15 lines of code
• 2 McCabe index (conditional complexity)

    def gen_imbalanced_data(self, img_num_per_cls):
        new_data = []
        new_targets = []
        targets_np = np.array(self.targets, dtype=np.int64)
        classes = np.unique(targets_np)

        self.num_per_cls_dict = dict()
        for the_class, the_img_num in zip(classes, img_num_per_cls):
            self.num_per_cls_dict[the_class] = the_img_num
            idx = np.where(targets_np == the_class)[0]
            # np.random.shuffle(idx) # This is very problametic. Different runs are using different training samples! So I removed this line.
            selec_idx = idx[:the_img_num]
            new_data.append(self.data[selec_idx, ...])
            new_targets.extend([the_class, ] * the_img_num)
        new_data = np.vstack(new_data)
        self.data = new_data
        self.targets = new_targets