renamed steps to step_size. step_size appears to be more intuitive.
fixed NumPy casting issues in PyTorchMixin's feature extraction method
added __post_init__ method in PyTorchMixin to define self.device = torch.device(self.device) during initialization of an object that inherits from PyTorchMixin rather than setting the device during every activation extraction (to me that seemed to be redundant computational overhead)
stepstostep_size.step_sizeappears to be more intuitive.PyTorchMixin's feature extraction method__post_init__method inPyTorchMixinto defineself.device = torch.device(self.device)during initialization of an object that inherits fromPyTorchMixinrather than setting the device during every activation extraction (to me that seemed to be redundant computational overhead)