Modelling Future Regressors
Future regressors are the external variables which have known future values. In that sense, the future regressors functionality if very similar to special events.
The past values of these regressors corresponding to the training time stamps, have to be provided along
with the training data itself. See below for an example where we create two dummy regressors A and B by
taking rolling means of the original data.
df['A'] = df['y'].rolling(7, min_periods=1).mean()
df['B'] = df['y'].rolling(30, min_periods=1).mean()
The dataframe created likewise, should look like below.
| ds | y | A | B | |
|---|---|---|---|---|
| 0 | 2007-12-10 | 9.59076 | 9.59076 | 9.59076 |
| 1 | 2007-12-11 | 8.51959 | 9.05518 | 9.05518 |
| 2 | 2007-12-12 | 8.18368 | 8.76468 | 8.76468 |
| 3 | 2007-12-13 | 8.07247 | 8.59162 | 8.59162 |
| 4 | 2007-12-14 | 7.89357 | 8.45201 | 8.45201 |
In order to perform forecasting, we also need to provide the future values of the regressors.
future_regressors_df = pd.DataFrame(data={'A': df['A'][:50], 'B': df['B'][:50]})
This dataframe looks like below.
| A | B | |
|---|---|---|
| 0 | 9.59076 | 9.59076 |
| 1 | 9.05518 | 9.05518 |
| 2 | 8.76468 | 8.76468 |
| 3 | 8.59162 | 8.59162 |
| 4 | 8.45201 | 8.45201 |
It is a dataframe with only the columns of the future values of the regressors.
Similar to events, future regressors too can be added in both the additive and multiplicative formats.
Additive Future Regressors
The default mode for future regressors in neural_prophet is additive. The regressors have to be added to
the NeuralProphet object by calling the add_future_regressor function. Once this is done, the model can be
fitted by providing to the fit function, the dataframe of the training data as well as the regressor values.
m = NeuralProphet(
n_forecasts=10,
yearly_seasonality=False,
weekly_seasonality=False,
daily_seasonality=False,
)
m = m.add_future_regressor(name='A')
m = m.add_future_regressor(name='B')
metrics = m.fit(df, freq="D")
When forecasting, the future dataframe must be created by providing the future values
of the regressors. To do that, now you need to call the make_future_dataframe function
by providing the previously created future_regressors_df as an argument.
future = m.make_future_dataframe(df=df, regressors_df=future_regressors_df, periods=3)
forecast = m.predict(df=future)
Now you can plot the components the same way as before and the resulting plot would look something like below.
fig_comp = m.plot_components(forecast)
In addition to the trend it also shows a plot for the additive future regressors. The coefficients of the future regressors can also be plotted.
fig_param = m.plot_parameters()
Multiplicative Future Regressors
Future regressors can also be added in multiplicative mode. You simply need to set
the mode to multiplicative when adding the regressors to the NeuralProphet object.
m = m.add_future_regressor(name='A', mode="multiplicative")
m = m.add_future_regressor(name='B')
In the above example, we have both additive and multiplicative regressors, where A
is multiplicative and B is additive. All the other steps in the fitting and the forecasting
processes are the same.
Regularization for Future Regressors
We can add regularization into the future regressors as below.
m = m.add_future_regressor(name='A', regularization=0.05)
m = m.add_future_regressor(name='B', regularization=0.02)
This will add sparsity into the individual regressor coefficients.