Training the model on all data about greykite

[KaixuYang]

Hi @Sigvesor what does your df_model look like? If the last available date in your df_model is 2020-2-20 9:00:00, then you don't have to explicitly specify the train_end_date. The forecast.forecast_result.forecast includes the forecasted values in the future period (starting 2020-2-20 10:00:00). And you can also use forecaster.model.predict on any future data frame similar to df_model.

The error looks like there are regressor columns in your model and the future values were not provided? Could you confirm that?

[Sigvesor]

Hi @KaixuYang, thanks for the reply.

You are correct, I have regressor columns. The df_model I give run_forecast_config() consist of the columns "ts", "y", "regressor1", "regressor2".

regressors = {
"regressor_cols": ["regressor1", "regressor2"]
}

All columns has the same last timestep 2020-2-20 9:00:00. I can confirm that future values are not provided. Am I supposed to give the future values in run_forecast_config() ? I though I would merge the future values of the regressors with the future dataframe and run forecaster.model.predict on this. This was how I used to proceed with fbprophet. What am I missing here?

[KaixuYang]

@Sigvesor When you run run_forecast_config, the method

• fits cross-validation
• fits train/test split
• fits model on entire data
• generates prediction for the future period (specified by forecast_horizon) based on the model fitted on entire data

So you don't have to call predict on yourself. If you include the future rows in the df_model, leaving the values to be NA and including the regressor future values, the forecast will be generated automatically and stored in forecast_result.forecast. And you can make forecast on different time period by calling the forecast_result.model.predict method.

[Sigvesor]

@KaixuYang,

Perfect, thanks for the help!