Fix/nlinear normalization for multivariate series (#2072)

* fix: properly slice x when normalizing the target features

* feat: updated changelog

CHANGELOG.md

@@ -30,6 +30,7 @@ but cannot always guarantee backwards compatibility. Changes that may **break co
 - Fixed a bug when using encoders with `RegressionModel` and series with a non-evenly spaced frequency (e.g. Month Begin). This raised an error during lagged data creation when trying to divide a pd.Timedelta by the ambiguous frequency. [#2034](https://github.com/unit8co/darts/pull/2034) by [Antoine Madrona](https://github.com/madtoinou).
 - Fixed a bug when loading a `TorchForecastingModel` that was trained with a precision other than `float64`. [#2046](https://github.com/unit8co/darts/pull/2046) by [Freddie Hsin-Fu Huang](https://github.com/Hsinfu).
 - Fixed broken links in the `Transfer learning` example notebook with publicly hosted version of the three datasets. [#2067](https://github.com/unit8co/darts/pull/2067) by [Antoine Madrona](https://github.com/madtoinou).
+- Fixed a bug when using `NLinearModel` on multivariate series with covariates and `normalize=True`. [#2072](https://github.com/unit8co/darts/pull/2072) by [Antoine Madrona](https://github.com/madtoinou).
 
 ### For developers of the library:
 

darts/models/forecasting/nlinear.py

@@ -144,7 +144,8 @@ def forward(
             if self.normalize:
                 # get last values only for target features
                 seq_last = x[:, -1:, : self.output_dim].detach()
-                x = x - seq_last
+                # normalize the target features only (ignore the covariates)
+                x[:, :, : self.output_dim] = x[:, :, : self.output_dim] - seq_last
 
             x = self.layer(x.view(batch, -1))  # (batch, out_len * out_dim * nr_params)
             x = x.view(
@@ -174,6 +175,7 @@ def forward(
 
             x = x.view(batch, self.output_chunk_length, self.output_dim, self.nr_params)
             if self.normalize:
+                # model only forecasts target components, no need to slice
                 x = x + seq_last.view(seq_last.shape + (1,))
         return x