add map to torch Loader

merlin/dataloader/torch.py

@@ -93,6 +93,17 @@ def __init__(
             transforms=transforms,
             device=device,
         )
+        self._map_fns = []
+
+    def map(self, fn):
+        """
+        Applying a function to each batch.
+
+        This can for instance be used to add `sample_weight` to the model.
+        """
+        self._map_fns.append(fn)
+
+        return self
 
     def __iter__(self):
         return LoaderBase.__iter__(self)
@@ -165,6 +176,14 @@ def _build_sparse_tensor(
             sparse_tensor = sparse_tensor.to_dense()
         return sparse_tensor
 
+    def _handle_tensors(self, tensors, tensor_names):
+        to_return = super()._handle_tensors(tensors, tensor_names)
+
+        for map_fn in self._map_fns:
+            to_return = map_fn(*to_return)
+
+        return to_return
+
     def _cast_to_numpy_dtype(self, dtype):
         """
         Get the numpy dtype from the framework dtype.

tests/unit/dataloader/test_torch_dataloader.py

@@ -328,3 +328,41 @@ def test_dataloader_schema(df, dataset, batch_size, cpu):
 
     num_label_cols = y.shape[1] if len(y.shape) > 1 else 1
     assert num_label_cols == 1
+
+
+def test_torch_map(tmpdir):
+    df = make_df(
+        {
+            "cat1": [1] * 100,
+            "cat2": [2] * 100,
+            "cat3": [3] * 100,
+            "label": [0] * 100,
+            "sample_weight": [1.0] * 100,
+            "cont2": [2.0] * 100,
+            "cont1": [1.0] * 100,
+        }
+    )
+    path = os.path.join(tmpdir, "dataset.parquet")
+    df.to_parquet(path)
+    ds = Dataset(df)
+    ds.schema["label"] = ds.schema["label"].with_tags(Tags.TARGET)
+
+    def add_sample_weight(features, labels, sample_weight_col_name="sample_weight"):
+        sample_weight = features.pop(sample_weight_col_name)
+
+        return features, labels, sample_weight
+
+    data_itr = torch_dataloader.Loader(
+        ds,
+        batch_size=10,
+        shuffle=False,
+    ).map(add_sample_weight)
+
+    for X, y, sample_weight in data_itr:
+        assert list(X["cat1"].cpu().numpy()) == [1] * 10
+        assert list(X["cat2"].cpu().numpy()) == [2] * 10
+        assert list(X["cat3"].cpu().numpy()) == [3] * 10
+        assert list(X["cont1"].cpu().numpy()) == [1.0] * 10
+        assert list(X["cont2"].cpu().numpy()) == [2.0] * 10
+
+        assert list(sample_weight.cpu().numpy()) == [1.0] * 10