Adjusted docs when needed

Author: jrzaurin

examples/scripts/mutil_tabular_components.py

@@ -20,10 +20,10 @@ class ModelFuser(BaseWDModelComponent):
         List of models whose outputs will be fused
     fusion_method: Union[str, List[str]]
         Method to fuse the output of the models. It can be one of
-        ['concatenate', 'mean', 'max', 'sum', 'mult', 'head'] or a list of
-        those. If a list is provided the output of the models will be fused
-        using all the methods in the list and the final output will be the
-        concatenation of the outputs of each method
+        ['concatenate', 'mean', 'max', 'sum', 'mult', 'dot', 'head'] or a
+        list of those, but 'dot'. If a list is provided the output of the
+        models will be fused using all the methods in the list and the final
+        output will be the concatenation of the outputs of each method
     projection_method: Optional[str]
         If the fusion_method is not 'concatenate', this parameter will
         determine how to project the output of the models to a common

pytorch_widedeep/models/model_fusion.py

@@ -53,7 +53,9 @@ class WideDeep(nn.Module):
     deeptabular: BaseWDModelComponent, Optional, default = None
         Currently this library implements a number of possible architectures
         for the `deeptabular` component. See the documenation of the
-        package.
+        package. Note that `deeptabular` can be a list of models. This is
+        useful when using multiple tabular inputs (e.g. for example in the
+        context of a two-tower model for recommendation systems)
     deeptext: BaseWDModelComponent | List[BaseWDModelComponent], Optional, default = None
         Currently this library implements a number of possible architectures
         for the `deeptext` component. See the documenation of the

pytorch_widedeep/models/wide_deep.py

@@ -47,7 +47,7 @@ def tabular_train_val_split(
     X_val: Optional[np.ndarray] = None,
     y_val: Optional[np.ndarray] = None,
     val_split: Optional[float] = None,
-):
+) -> Tuple[TensorDataset, Optional[TensorDataset]]:
     r"""
     Function to create the train/val split for the BayesianTrainer where only
     tabular data is used
@@ -86,7 +86,7 @@ def tabular_train_val_split(
             torch.from_numpy(y_val),
         )
     elif val_split is not None:
-        y_tr, y_val, idx_tr, idx_val = train_test_split(
+        y_tr, y_val, idx_tr, idx_val = train_test_split(  # type: ignore
             y,
             np.arange(len(y)),
             test_size=val_split,
@@ -140,13 +140,13 @@ def wd_train_val_split(  # noqa: C901
         random seed to be used during train/val split
     method: str
         'regression',  'binary' or 'multiclass'
-    X_wide: np.ndarray, Optional, default = None
+    X_wide: np.ndaaray, Optional, default = None
         wide dataset
-    X_tab: np.ndarray, Optional, default = None
+    X_tab: np.ndarray or List[np.ndarray], Optional, default = None
         tabular dataset (categorical and continuous features)
-    X_img: np.ndarray, Optional, default = None
+    X_img: np.ndarray or List[np.ndarray], Optional, default = None
         image dataset
-    X_text: np.ndarray, Optional, default = None
+    X_text: np.ndarray or List[np.ndarray], Optional, default = None
         text dataset
     X_val: Dict, Optional, default = None
         Dict with the validation set, where the keys are the component names

pytorch_widedeep/training/_trainer_utils.py

@@ -17,7 +17,7 @@ class WideDeepDataset(Dataset):
     ----------
     X_wide: np.ndarray
         wide input
-    X_tab: np.ndarray
+    X_tab: np.ndarray or List[np.ndarray]
         deeptabular input
     X_text: np.ndarray or List[np.ndarray]
         deeptext input

pytorch_widedeep/training/_wd_dataset.py

@@ -298,7 +298,9 @@ def fit(  # noqa: C901
             See `pytorch_widedeep.preprocessing.WidePreprocessor`
         X_tab: np.ndarray, Optional. default=None
             Input for the `deeptabular` model component.
-            See `pytorch_widedeep.preprocessing.TabPreprocessor`
+            See `pytorch_widedeep.preprocessing.TabPreprocessor`. If multiple
+            tabular models are used for different columns, this should be a
+            list of numpy arrays
         X_text: Union[np.ndarray, List[np.ndarray]], Optional. default=None
             Input for the `deeptext` model component.
             See `pytorch_widedeep.preprocessing.TextPreprocessor`.
@@ -547,13 +549,13 @@ def predict(  # type: ignore[override, return]
         X_wide: np.ndarray, Optional. default=None
             Input for the `wide` model component.
             See `pytorch_widedeep.preprocessing.WidePreprocessor`
-        X_tab: np.ndarray, Optional. default=None
+        X_tab: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptabular` model component.
             See `pytorch_widedeep.preprocessing.TabPreprocessor`
-        X_text: np.ndarray, Optional. default=None
+        X_text: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptext` model component.
             See `pytorch_widedeep.preprocessing.TextPreprocessor`
-        X_img: np.ndarray, Optional. default=None
+        X_img: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deepimage` model component.
             See `pytorch_widedeep.preprocessing.ImagePreprocessor`
         X_test: Dict, Optional. default=None
@@ -606,13 +608,13 @@ def predict_uncertainty(  # type: ignore[return]
         X_wide: np.ndarray, Optional. default=None
             Input for the `wide` model component.
             See `pytorch_widedeep.preprocessing.WidePreprocessor`
-        X_tab: np.ndarray, Optional. default=None
+        X_tab: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptabular` model component.
             See `pytorch_widedeep.preprocessing.TabPreprocessor`
-        X_text: np.ndarray, Optional. default=None
+        X_text: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptext` model component.
             See `pytorch_widedeep.preprocessing.TextPreprocessor`
-        X_img: np.ndarray, Optional. default=None
+        X_img: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deepimage` model component.
             See `pytorch_widedeep.preprocessing.ImagePreprocessor`
         X_test: Dict, Optional. default=None
@@ -700,13 +702,13 @@ def predict_proba(  # type: ignore[override, return]  # noqa: C901
         X_wide: np.ndarray, Optional. default=None
             Input for the `wide` model component.
             See `pytorch_widedeep.preprocessing.WidePreprocessor`
-        X_tab: np.ndarray, Optional. default=None
+        X_tab: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptabular` model component.
             See `pytorch_widedeep.preprocessing.TabPreprocessor`
-        X_text: np.ndarray, Optional. default=None
+        X_text: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deeptext` model component.
             See `pytorch_widedeep.preprocessing.TextPreprocessor`
-        X_img: np.ndarray, Optional. default=None
+        X_img: np.ndarray or List[np.ndarray], Optional. default=None
             Input for the `deepimage` model component.
             See `pytorch_widedeep.preprocessing.ImagePreprocessor`
         X_test: Dict, Optional. default=None