Why does the input window and output window overlap

[stoyanoronaldo]

Hello again @lingbai-kong,

In the TimeseriesdDataLoader:

for i in range(self.time_step,len(self.data)+1):
self.dataset.append((self.data[i-self.time_step:i].reshape(self.time_step,self.series_num,self.feature_dim) ,self.data[i-self.output_window:i].reshape(self.output_window,self.series_num,self.output_dim)))

why does the input window and output window overlap, if I'm getting the things right?

Best regards

[lingbai-kong]

Hello, “the input window and output window overlaps" is in line with expectations. Maybe you concern this could lead to label leak problem because it is quiet unusual for a forecasting model. However, thanks to the multi-kernel causal convolution block, when the model forecasts certain value in output window, it will not get any information after the time point of that value. So label leak problem does not happen.
@stoyanoronaldo