Problematic shake-shake and shake-drop implementation

[ArlindKadra]

Auto-PyTorch/autoPyTorch/pipeline/components/setup/network_backbone/utils.py

Lines 97 to 99 in 9002937

	if is_training:
	result = (torch.FloatTensor([0.5]), torch.FloatTensor([0.5]))
	return result if not is_cuda else (result[0].cuda(), result[1].cuda())

Shake-Shake and Shake-Drop do not get an alpha and beta that is sampled during training time and fixed during test time, the intended behavior is reversed.

[ravinkohli]

this has been fixed in #249.