Adding Future Regressor Error

[Msaleh87]

Discussed in #797

^{Originally posted by Msaleh87 October 3, 2022}
So I prepared my data as shown below:

This is my code, L12 is the dataset you see in the pic above:

L12M = L_prophet_5min_T[0:(12*30*24*12)] # the dataset show in the above pic
m = NeuralProphet(n_forecasts=(2*24*12),
    n_lags=(16*24*12),
    yearly_seasonality="auto",
    weekly_seasonality="auto",  
    daily_seasonality="auto",
    growth="off", seasonality_reg= 0.5, optimizer = 'AdamW',
    learning_rate = 0.1, n_changepoints="auto", changepoints_range = "auto",  batch_size = 16 ) 

m.add_country_holidays("US", mode="additive") 
m = m.add_future_regressor(name='A')
m = m.add_future_regressor(name='B')
m = m.add_future_regressor(name='C')
metrics = m.fit(L12M, freq="5min") 
m.highlight_nth_step_ahead_of_each_forecast(step_number=m.n_forecasts)

# the future values of A, B, and C
future_regressors_df = pd.DataFrame(data={'A': L_prophet_5min_T['A'][(12*30*24*12):(12*30*24*12)+576], 
                                          'B': L_prophet_5min_T['B'][(12*30*24*12):(12*30*24*12)+576],
                                          'C': L_prophet_5min_T['C'][(12*30*24*12):(12*30*24*12)+576]})

The model trains and saves fine, then when I try to forecast using this code, I get an error:

P_horizon = 576
future = m.make_future_dataframe(df = L12M, periods = P_horizon, n_historic_predictions=True) 
forecast1 = m.predict(df=future)

Should I add an argument for the future regressors inside the forecast?!

[ourownstory]

Should I add an argument for the future regressors inside the forecast?!

Their future values should be added in make_future_dataframe, see:

neural_prophet/neuralprophet/forecaster.py

Line 1164 in 3bf0336

regressor_df : pd.DataFrame

Hope this helps

[Msaleh87]

Thanks @ourownstory, it worked

So I needed to have the values of the future regressors with index starts from 0 and I needed to regressor_df to make_future_dataframe as you said. I will leave the code here for future reference:

This is how I sat up the future_regressors_df

future_regressors_df = pd.DataFrame(data={'A': L_prophet_5min_T['A'][(12*30*24*12):(12*30*24*12)+(288*2)], 
                                          'B': L_prophet_5min_T['B'][(12*30*24*12):(12*30*24*12)+(288*2)],
                                          'C': L_prophet_5min_T['C'][(12*30*24*12):(12*30*24*12)+(288*2)]})
# To make sure the indexing starts from 0, I did this:
future_regressors_df = future_regressors_df.reset_index() 
future_regressors_df = future_regressors_df.drop(['index'], axis=1)
future_regressors_df

Then during the forecast:

P_horizon = 576
future = m.make_future_dataframe(df = L12M, **regressors_df=future_regressors_df** , periods = P_horizon, n_historic_predictions=576) 
#def make_future_dataframe(self, df, events_df=None, regressors_df=None, periods=None, n_historic_predictions=False):

forecast1 = m.predict(df=future)

[ourownstory]

@Msaleh87 Thank you for following-up on this!
did it fail if you did not reset the index?
If so, please open an issue, we can drop the index internally instead.