I am currently using NeuralForecast using historical and future exogenous variables. The list of models works fine in the case of historical exogenous variables; however, it does not work when I pass future covariates. Am I missing something?

Hi @Yalemzewd Nigussie. We have this tutorial on exogenous variables: https://nixtla.github.io/neuralforecast/examples/exogenous_variables.html.

There is no error message. But, forecasts are NAN

Are you using the

predict

method? If so, you also need to pass the

futr_df

dataset with information for the future variables in the forecasting window (

h

values per unique_id). We are going to add a protection to prevent this from happening and give an informative error message.

image.png

futr_df

should not be a list, but a pandas dataframe with the values of the exogenous variables. Here is an example in the tutorial I sent you

Sorry, it was like this: Y_hat_df = nf1.predict(futr_df=df_future).reset_index()

This weekly data right? can you set the frequency to 'W-SUN'?

OK, let me try and get back to you!

The most common issue with futr variables is some mismatch in the

futr_df

dataframe. Maybe: 1. A variable in the

futr_exog_list

is not in

futr_df

(wrong spelling for example). 2. Mismatch in future dates. The number of rows for each unique_id should match

h

exactly. In the current branch we have a protection for this case (not in pip yet)

Good Catch. Thanks, let me check that again

I suggest you to start with a simpler case. Maybe only limit to your first unique_id and only one futr_variable, and test it. Then add more unique_ids, etc. It will be easier to detect the error.

Thank Cris for your prompt response and help to identify the issue I had