Thanks for sharing this code, and I have a question with it.

When I run this code on Colab with the original setup，it would run perfectly. But when I change the sequence_length = 4 and step_size = 2 (just like the description in With sequence_length of 4 and step_size of 2, we end up with 498,623 sequences.), the code has an error with
Sizes of tensors must match except in dimension 2. Expected size 4 but got size 8 for tensor number 1 in the list.
in line transfomer_features = torch.cat((transfomer_features, positional_embedding), dim=2)
What should I do to get it to work properly?

P.S. Sorry for my poor English.

Hello sir, why is test_step part missing?

Is this for an unknown user or an unknown movie?

Thanks a lot.

Obviously, only fixed-length sequences are used here, and there is no filling operation, which limits the scope of application of this model.

Helo，thanks for your complete code. I read this code carefully，and have a question.

It's BST model's forward function

transfomer_features, user_features, target_movie_rating = self.encode_input(batch)
transfomer_features = self.positional_embedding(transfomer_features)

And it's positional_embedding module

class PositionalEmbedding(nn.Module):
    def __init__(self, max_len, d_model):
        super().__init__()

        self.pe = nn.Embedding(max_len, d_model)

    def forward(self, x):
        batch_size = x.size(0)
        return self.pe.weight.unsqueeze(0).repeat(batch_size, 1, 1)

we can see that PositionalEmbedding do nothing, just return it's weight. PositionalEmbedding doesn't have any interaction with x. It's really confused.

The forward function of the TransformerEncoderLayer can have src_key_padding_mask. Maybe we can update it too.

I wonder where 661 comes from.

thanks.