def training_step(self, batch, batch_idx):
fc, (mag_min, mag_max) = batch
x_fc = fc[:, self.dst_flatten_order][:, :-1]
y_fc = fc[:, self.dst_flatten_order][:, 1:]
pred = self.sres.forward(x_fc)
fc_loss, amp_loss, phi_loss = self.criterion(pred, y_fc, mag_min, mag_max)
return {'loss': fc_loss, 'amp_loss': amp_loss, 'phi_loss': phi_loss}
These are your functions in the class named SResTransformerModule. When I debug it, I found that fc.shape is [16, 2016, 2]. Therefore, x_fc[:, 1:, :] will be equal to y_pc[:, :-1, :], which means the input of the network is the same as the output of the network. Is this a mistake, or is there something I don't understand?
Thank you for your help!
These are your functions in the class named SResTransformerModule. When I debug it, I found that fc.shape is [16, 2016, 2]. Therefore, x_fc[:, 1:, :] will be equal to y_pc[:, :-1, :], which means the input of the network is the same as the output of the network. Is this a mistake, or is there something I don't understand?
Thank you for your help!