Backports for v0.16.2 (#3260)

*Description of changes:* backport changes
- #3259 
- #3261

By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.


**Please tag this pr with at least one of these labels to make our
release process faster:** BREAKING, new feature, bug fix, other change,
dev setup

---------

Co-authored-by: Abdul Fatir <[email protected]>

Author: lostella

requirements/requirements-extras-r.txt

@@ -1 +1 @@
-rpy2~=3.5
+rpy2~=3.5,<3.6

requirements/requirements-extras-statsforecast.txt

@@ -1 +1,4 @@
+# scipy cap can be removed once this is resolved: https://github.com/statsmodels/statsmodels/issues/9584
+scipy<1.16.0; python_version > "3.7.0"
+scipy~=1.7.3; python_version <= "3.7.0"
 statsforecast~=1.0

requirements/requirements-pytorch.txt

@@ -2,5 +2,6 @@ torch>=1.9,<3
 lightning>=2.2.2,<2.5
 # Capping `lightning` does not cap `pytorch_lightning`, so we cap manually
 pytorch_lightning>=2.2.2,<2.5
-scipy~=1.10; python_version > "3.7.0"
+# scipy cap can be removed once this is resolved: https://github.com/statsmodels/statsmodels/issues/9584
+scipy<1.16.0; python_version > "3.7.0"
 scipy~=1.7.3; python_version <= "3.7.0"

src/gluonts/transform/split.py

@@ -494,7 +494,7 @@ def __init__(
         is_pad_field: str = FieldName.IS_PAD,
         start_field: str = FieldName.START,
         forecast_start_field: str = FieldName.FORECAST_START,
-        observed_value_field: str = FieldName.OBSERVED_VALUES,
+        observed_value_field: Optional[str] = FieldName.OBSERVED_VALUES,
         lead_time: int = 0,
         output_NTC: bool = True,
         time_series_fields: List[str] = [],
@@ -529,11 +529,9 @@ def flatmap_transform(
 
         sampled_indices = self.instance_sampler(target)
 
-        slice_cols = (
-            self.ts_fields
-            + self.past_ts_fields
-            + [self.target_field, self.observed_value_field]
-        )
+        slice_cols = self.ts_fields + self.past_ts_fields + [self.target_field]
+        if self.observed_value_field is not None:
+            slice_cols.append(self.observed_value_field)
         for i in sampled_indices:
             pad_length = max(self.past_length - i, 0)
             d = data.copy()