Feat/tft static categorical

.github/workflows/merge.yml

@@ -87,7 +87,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        example-name: [00-quickstart.ipynb, 01-multi-time-series-and-covariates.ipynb, 02-data-processing.ipynb, 03-FFT-examples.ipynb, 04-RNN-examples.ipynb, 05-TCN-examples.ipynb, 06-Transformer-examples.ipynb, 07-NBEATS-examples.ipynb, 08-DeepAR-examples.ipynb, 09-DeepTCN-examples.ipynb, 10-Kalman-filter-examples.ipynb, 11-GP-filter-examples.ipynb, 12-Dynamic-Time-Warping-example.ipynb, 13-TFT-examples.ipynb]
+        example-name: [00-quickstart.ipynb, 01-multi-time-series-and-covariates.ipynb, 02-data-processing.ipynb, 03-FFT-examples.ipynb, 04-RNN-examples.ipynb, 05-TCN-examples.ipynb, 06-Transformer-examples.ipynb, 07-NBEATS-examples.ipynb, 08-DeepAR-examples.ipynb, 09-DeepTCN-examples.ipynb, 10-Kalman-filter-examples.ipynb, 11-GP-filter-examples.ipynb, 12-Dynamic-Time-Warping-example.ipynb, 13-TFT-examples.ipynb, 15-static-covariates.ipynb]
     steps:
       - name: "1. Clone repository"
         uses: actions/checkout@v2

darts/dataprocessing/transformers/__init__.py

@@ -15,3 +15,4 @@
     TopDownReconciliator,
 )
 from .scaler import Scaler
+from .static_covariates_transformer import StaticCovariatesTransformer

darts/models/forecasting/block_rnn_model.py

@@ -325,6 +325,10 @@ def __init__(
         self.n_rnn_layers = n_rnn_layers
         self.dropout = dropout
 
+    @staticmethod
+    def _supports_static_covariates() -> bool:
+        return False
+
     def _create_model(self, train_sample: Tuple[torch.Tensor]) -> torch.nn.Module:
         # samples are made of (past_target, past_covariates, future_target)
         input_dim = train_sample[0].shape[1] + (

darts/models/forecasting/forecasting_model.py

@@ -104,6 +104,10 @@ def __init__(self, *args, **kwargs):
         # This is only used if the model has been fit on one time series.
         self.training_series: Optional[TimeSeries] = None
 
+        # Static covariates sample from the (first) target series used for training the model through the `fit()`
+        # function.
+        self.static_covariates: Optional[pd.DataFrame] = None
+
         # state; whether the model has been fit (on a single time series)
         self._fit_called = False
 
@@ -959,11 +963,13 @@ def fit(
         if isinstance(series, TimeSeries):
             # if only one series is provided, save it for prediction time (including covariates, if available)
             self.training_series = series
+            self.static_covariates = series.static_covariates
             if past_covariates is not None:
                 self.past_covariate_series = past_covariates
             if future_covariates is not None:
                 self.future_covariate_series = future_covariates
         else:
+            self.static_covariates = series[0].static_covariates
             if past_covariates is not None:
                 self._expect_past_covariates = True
             if future_covariates is not None:

darts/models/forecasting/nbeats.py

@@ -762,6 +762,10 @@ def __init__(
         if isinstance(layer_widths, int):
             self.layer_widths = [layer_widths] * num_stacks
 
+    @staticmethod
+    def _supports_static_covariates() -> bool:
+        return False
+
     def _create_model(self, train_sample: Tuple[torch.Tensor]) -> torch.nn.Module:
         # samples are made of (past_target, past_covariates, future_target)
         input_dim = train_sample[0].shape[1] + (

darts/models/forecasting/nhits.py

@@ -757,6 +757,10 @@ def _check_sizes(tup, name):
 
         return pooling_kernel_sizes, n_freq_downsample
 
+    @staticmethod
+    def _supports_static_covariates() -> bool:
+        return False
+
     def _create_model(self, train_sample: Tuple[torch.Tensor]) -> torch.nn.Module:
         # samples are made of (past_target, past_covariates, future_target)
         input_dim = train_sample[0].shape[1] + (

darts/models/forecasting/rnn_model.py

@@ -466,3 +466,7 @@ def _verify_train_dataset_type(self, train_dataset: TrainingDataset):
             train_dataset.ds_past.shift == 1,
             "RNNModel requires a shifted training dataset with shift=1.",
         )
+
+    @staticmethod
+    def _supports_static_covariates() -> bool:
+        return False

darts/models/forecasting/tcn_model.py

@@ -490,3 +490,7 @@ def _build_train_dataset(
             shift=self.output_chunk_length,
             max_samples_per_ts=max_samples_per_ts,
         )
+
+    @staticmethod
+    def _supports_static_covariates() -> bool:
+        return False