Difference between past_target、feat_dynamic_real and past_feat_dynamic_real

[huyz1117]

Thank you very much for your excellent work！

I am using Moirai for electricity load forecasting, and during inference, I noticed that the forward function receives inputs such as past_target、 feat_dynamic_real、and past_feat_dynamic_real. Could you please explain what these three parameters represent?

For example, if I want to use the historical load data from the past 7 days, historical weather data (temperature, humidity, irradiance, etc.), forecasted weather data for the next 2 days, and date types (holiday information, hour, minute, etc.) as features to predict the load for the next 2 days, how should I handle these features?Which parameter do they belong to respectively?

    def forward(
        self,
        past_target: Float[torch.Tensor, "batch past_time tgt"],
        past_observed_target: Bool[torch.Tensor, "batch past_time tgt"],
        past_is_pad: Bool[torch.Tensor, "batch past_time"],
        feat_dynamic_real: Optional[Float[torch.Tensor, "batch time feat"]] = None,
        observed_feat_dynamic_real: Optional[
            Float[torch.Tensor, "batch time feat"]
        ] = None,
        past_feat_dynamic_real: Optional[
            Float[torch.Tensor, "batch past_time past_feat"]
        ] = None,
        past_observed_feat_dynamic_real: Optional[
            Float[torch.Tensor, "batch past_time past_feat"]
        ] = None,
        num_samples: Optional[int] = None,
    ) -> Float[torch.Tensor, "batch sample future_time *tgt"]: