a question on the input and output values

[lk1983823]

Thank you for the great work. But when I debugging the example file, I find that the last 24 rows of target (power_usage) in data_input have the same value of the that in the data_output, where the shape of the data_input is (batch_size, 192, 5) and the data_output is (batch_size, 24,1). This can be printed in the def __getitem__(self, idx) of class TFTDataset. Obviously, the input should not include the value in output. And the proper shape of inputs is 192-24 = 168, which exclude the power_usage of the last 24 values, in my opinion. What is the problem? Thanks.

[Tc511]

Thank you for the great work. But when I debugging the example file, I find that the last 24 rows of target (power_usage) in data_input have the same value of the that in the data_output, where the shape of the data_input is (batch_size, 192, 5) and the data_output is (batch_size, 24,1). This can be printed in the def __getitem__(self, idx) of class TFTDataset. Obviously, the input should not include the value in output. And the proper shape of inputs is 192-24 = 168, which exclude the power_usage of the last 24 values, in my opinion. What is the problem? Thanks.

    # Isolate known and observed historical inputs.
    if unknown_inputs is not None:
          historical_inputs = torch.cat([
              unknown_inputs[:, :self.num_encoder_steps, :],
              known_combined_layer[:, :self.num_encoder_steps, :],
              obs_inputs[:, :self.num_encoder_steps, :]
          ], dim=-1)
    else:
          historical_inputs = torch.cat([
              known_combined_layer[:, :self.num_encoder_steps, :],
              obs_inputs[:, :self.num_encoder_steps, :]
          ], dim=-1)
          
          Here only the first 168 of the input data are taken