social-lstm
social-lstm copied to clipboard
Loss function computation
Hi, I have a question. In training process:"
Forward prop
outputs, _, _ = net(x_seq, grid_seq, hidden_states, cell_states, PedsList_seq,numPedsList_seq ,dataloader, lookup_seq)
Compute loss
loss = Gaussian2DLikelihood(outputs, x_seq, PedsList_seq, lookup_seq)"
why the loss is computed by using x_seq and outputs rather than outputs and y_seq? Thanks
x_seq represents the observed part of a trajectory and y_seq is unknown part (will be predicted). During the training, we are using known (observed) part of a trajectory for the training model and calculating training loss.
@quancore hello. I am also confused of the loss calculation. In that case, do you mean that the model mainly for learning the sequence relationship among the hidden state, so it doesn't matter weather its output was compare with the known (observed) part or unknown part? But will it infect the model performance to predict the unknown situation as it was always trained to predict the known part? Thank you and waiting for your reply.