All parameters should be initialized the same time when calling reset_parameters()
px_r is the only torch.nn.Parameters rather than a neural network. We could initialize px_r in weight_init. However, there are 3 options of initialization of px_r depending on the setup of dispersion. Moving initialization of px_r into weight_init makes the choice of initialization of px_r inflexible.
All parameters should be initialized the same time when calling reset_parameters()
px_r is the only torch.nn.Parameters rather than a neural network. We could initialize px_r in weight_init. However, there are 3 options of initialization of px_r depending on the setup of dispersion. Moving initialization of px_r into weight_init makes the choice of initialization of px_r inflexible.