Some personal questions

[719304040]

I am eager to integrate this work into my own project. However, I have a question regarding certain parts of the code and would greatly appreciate the author's assistance.

When calculating the logits of the Q-values (in def get_q_values), why is torch.roll used to shift embedding_values? I am struggling to understand why this step is necessary, as it seems to only occur during backpropagation.

action_bin_embeddings = self.action_bin_embeddings[:num_actions] action_bin_embeddings = torch.roll(action_bin_embeddings, shifts = -1, dims = 1) logits = einsum('b n d, n a d -> b n a', embed, action_bin_embeddings)