I am using Informer in the Transformer library for time series forecasting of weather variable.
The model has parameters like: past_time_features, future_values, past_values, future_time_features.
During training,
the data has access to all these parameters.
past_time_features: day of year , weather predictors (like rain, temper), etc.
future_time_features: day of year , weather predictors (like rain, temper), etc. (during the forecasting period)
However, during the inference we will not have access to the weather predictors which are dynamic and temporal features.
In that case, how do we use the model to make sure we get the forecasting during the inference? Because if I don’t have the feature_time_features during the inference, the model will prompt an error asking for it.
Questions:
- Are these variables like precip, temperature,etc passed to the past_time_features or past_values?
- If passed to past_time_features and future_time_features, what to do during the inference time as the future will not have access to these variables to pass to the model?
Prediction:
Forecast snow-melt value using all the historical weather features.