N-BEATS block layer indexing

[respwill]

def call(self, inputs):
          x = inputs
          for layer in self.hidden:
              x = layer(x)
          theta = self.theta_layer(x)

          """
          theta layer is split by input_size and horizon_size to crate backcast and forecast,
          I read paper but couldn't find sentence about how to set index to separate theta into backcast and forecast
          Could anyone show me the paragraph to explain about it in "N-BEATS: NEURAL BASIS EXPANSION ANALYSIS FOR
          INTERPRETABLE TIME SERIES FORECASTING"?
          """

          backcast, forecast = theta[:, :self.input_size], theta[:, -self.horizon:]  #Question point!
          return backcast, forecast

*I added my main question as comments in the above code.
how to know self.input_size, -self.horizon is proper index to separate theta into backcast and forecast?

[mrdbourke]

Hey @respwill,

self.input_size is the number of datapoints used to predict the horizon.

Hence the backcast (forecasting backwards) is the same size as the input_size and the forecast is the same size as the horizon.