Closed xiaoqi91 closed 1 year ago
@xiaoqi91 Sure, it is okay as long as you call the init_hidden function before the Kalman filtering iteration.
@xiaoqi91 Sure, it is okay as long as you call the init_hidden function before the Kalman filtering iteration.
thank you for your reply! one more question, there is only one complex gru layer in file src/nkf.py, but the paper writes that the model consists of one complex fully-connected (FC) layer, two complex GRU layers, and another two complex FC layers. they seem to be inconsistent. . .could you please explain this question?
@xiaoqi91 Oops, it should be a typo. Only one complex GRU layer is used in the paper.
@xiaoqi91 Oops, it should be a typo. Only one complex GRU layer is used in the paper.
ok, thank you !
could you please explain that why did you put the init step "self.kg_net.init_hidden(B * F, device)" in the forward function of class NKF? This will put gru weights to be zero, is it OK when training ?