[DOC] Add usage examples for TimeXer and TemporalFusionTransformer

pytorch_forecasting/models/temporal_fusion_transformer/_tft.py

@@ -125,6 +125,53 @@ class TemporalFusionTransformer(BaseModelWithCovariates):
         Bias for the mask in ScaledDotProductAttention.forward, by default -1e9.
         Set to -float("inf") to allow mixed precision training.
     **kwargs: additional arguments to :py:class:`~BaseModel`.
+
+    Examples
+    --------
+    >>> import pandas as pd
+    >>> import numpy as np
+    >>> from lightning.pytorch import Trainer
+    >>> from pytorch_forecasting import TimeSeriesDataSet
+    >>> from pytorch_forecasting.models import TemporalFusionTransformer
+    >>>
+    >>> # Generate toy sequential data
+    >>> data = pd.DataFrame(
+    ...     {
+    ...         "time_idx": np.tile(np.arange(40), 2),
+    ...         "target": np.random.randn(80),
+    ...         "group": np.repeat(["A", "B"], 40),
+    ...     }
+    ... )
+    >>>
+    >>> training = TimeSeriesDataSet(
+    ...     data,
+    ...     time_idx="time_idx",
+    ...     target="target",
+    ...     group_ids=["group"],
+    ...     min_encoder_length=10,
+    ...     max_encoder_length=10,
+    ...     min_prediction_length=5,
+    ...     max_prediction_length=5,
+    ...     time_varying_unknown_reals=["target"],
+    ... )
+    >>> dataloader = training.to_dataloader(batch_size=4)
+    >>>
+    >>> model = TemporalFusionTransformer.from_dataset(
+    ...     training,
+    ...     hidden_size=8,
+    ...     attention_head_size=1,
+    ...     hidden_continuous_size=4,
+    ...     dropout=0.1,
+    ... )
+    >>> trainer = Trainer(
+    ...     fast_dev_run=True,
+    ...     accelerator="cpu",
+    ...     enable_model_summary=False,
+    ...     enable_progress_bar=False,
+    ...     logger=False,
+    ... )
+    >>> trainer.fit(model, dataloader)
+    >>> predictions = model.predict(dataloader)
     """  # noqa: E501
 
     @classmethod

pytorch_forecasting/models/timexer/_timexer.py

@@ -38,7 +38,56 @@
 
 
 class TimeXer(BaseModelWithCovariates):
-    """TimeXer model for time series forecasting with exogenous variables."""
+    """TimeXer model for time series forecasting with exogenous variables.
+
+    Examples
+    --------
+    >>> import pandas as pd
+    >>> import numpy as np
+    >>> from lightning.pytorch import Trainer
+    >>> from pytorch_forecasting import TimeSeriesDataSet
+    >>> from pytorch_forecasting.models import TimeXer
+    >>>
+    >>> # Generate toy sequential data
+    >>> data = pd.DataFrame(
+    ...     {
+    ...         "time_idx": np.tile(np.arange(40), 2),
+    ...         "target": np.random.randn(80),
+    ...         "group": np.repeat(["A", "B"], 40),
+    ...     }
+    ... )
+    >>>
+    >>> training = TimeSeriesDataSet(
+    ...     data,
+    ...     time_idx="time_idx",
+    ...     target="target",
+    ...     group_ids=["group"],
+    ...     min_encoder_length=16,
+    ...     max_encoder_length=16,
+    ...     min_prediction_length=4,
+    ...     max_prediction_length=4,
+    ...     time_varying_unknown_reals=["target"],
+    ... )
+    >>> dataloader = training.to_dataloader(batch_size=4)
+    >>>
+    >>> model = TimeXer.from_dataset(
+    ...     training,
+    ...     hidden_size=16,
+    ...     n_heads=2,
+    ...     e_layers=1,
+    ...     d_ff=32,
+    ...     patch_length=8,
+    ... )
+    >>> trainer = Trainer(
+    ...     fast_dev_run=True,
+    ...     accelerator="cpu",
+    ...     enable_model_summary=False,
+    ...     enable_progress_bar=False,
+    ...     logger=False,
+    ... )
+    >>> trainer.fit(model, dataloader)
+    >>> predictions = model.predict(dataloader)
+    """
 
     @classmethod
     def _pkg(cls):