fix(tensor_distribution): replace RuntimeError with ValueError for parameter validation

Replace RuntimeError with more semantically appropriate ValueError for parameter validation errors across tensor distribution classes.

Author: Tim Joseph

src/tensorcontainer/tensor_distribution/categorical.py

@@ -28,7 +28,7 @@ def __init__(
         # check here to safely derive shape and device from the data tensor
         # before calling the parent constructor
         if data is None:
-            raise RuntimeError("Either 'probs' or 'logits' must be provided.")
+            raise ValueError("Either 'probs' or 'logits' must be provided.")
 
         # Store the parameters in annotated attributes before calling super().__init__()
         # This is required because super().__init__() calls self.dist() which needs these attributes

src/tensorcontainer/tensor_distribution/multinomial.py

@@ -25,7 +25,7 @@ def __init__(
     ):
         data = probs if probs is not None else logits
         if data is None:
-            raise RuntimeError("Either 'probs' or 'logits' must be provided.")
+            raise ValueError("Either 'probs' or 'logits' must be provided.")
 
         self._total_count = total_count
         self._probs = probs
@@ -83,4 +83,4 @@ def param_shape(self) -> Size:
         elif self._logits is not None:
             return self._logits.shape
         else:
-            raise RuntimeError("Neither probs nor logits are available.")
+            raise ValueError("Neither probs nor logits are available.")

src/tensorcontainer/tensor_distribution/one_hot_categorical.py

@@ -26,7 +26,7 @@ def __init__(
         # check here to safely derive shape and device from the data tensor
         # before calling the parent constructor
         if data is None:
-            raise RuntimeError("Either 'probs' or 'logits' must be provided.")
+            raise ValueError("Either 'probs' or 'logits' must be provided.")
 
         # Store the parameters in annotated attributes before calling super().__init__()
         # This is required because super().__init__() calls self.dist() which needs these attributes

src/tensorcontainer/tensor_distribution/relaxed_one_hot_categorical.py

@@ -27,13 +27,13 @@ def __init__(
         That is why we only allowed scalar temperatures for now.
         """
         if temperature.ndim > 0:
-            raise RuntimeError(
+            raise ValueError(
                 "Expected scalar temperature tensor. This is because of a bug in torch: https://github.com/pytorch/pytorch/issues/37162"
             )
 
         data = probs if probs is not None else logits
         if data is None:
-            raise RuntimeError("Either 'probs' or 'logits' must be provided.")
+            raise ValueError("Either 'probs' or 'logits' must be provided.")
 
         # Determine shape and device from data (probs or logits)
         shape = data.shape[:-1]

tests/tensor_distribution/test_categorical.py

@@ -15,7 +15,7 @@ class TestTensorCategoricalInitialization:
     def test_init_no_params_raises_error(self):
         """A ValueError should be raised when neither probs nor logits are provided."""
         with pytest.raises(
-            RuntimeError, match="Either 'probs' or 'logits' must be provided."
+            ValueError, match="Either 'probs' or 'logits' must be provided."
         ):
             TensorCategorical()
 

tests/tensor_distribution/test_one_hot_categorical.py

@@ -17,7 +17,7 @@ class TestTensorOneHotCategoricalInitialization:
     def test_init_no_params_raises_error(self):
         """A ValueError should be raised when neither probs nor logits are provided."""
         with pytest.raises(
-            RuntimeError, match="Either 'probs' or 'logits' must be provided."
+            ValueError, match="Either 'probs' or 'logits' must be provided."
         ):
             TensorOneHotCategorical()
 

tests/tensor_distribution/test_one_hot_categorical_straight_through.py

@@ -17,7 +17,7 @@ class TestTensorOneHotCategoricalStraightThroughInitialization:
     def test_init_no_params_raises_error(self):
         """A ValueError should be raised when neither probs nor logits are provided."""
         with pytest.raises(
-            RuntimeError, match="Either 'probs' or 'logits' must be provided."
+            ValueError, match="Either 'probs' or 'logits' must be provided."
         ):
             TensorOneHotCategoricalStraightThrough()
 

tests/tensor_distribution/test_relaxed_one_hot_categorical.py

@@ -45,7 +45,7 @@ def test_valid_initialization(
     def test_init_no_params_raises_error(self):
         """A RuntimeError should be raised when neither probs nor logits are provided."""
         with pytest.raises(
-            RuntimeError, match="Either 'probs' or 'logits' must be provided."
+            ValueError, match="Either 'probs' or 'logits' must be provided."
         ):
             TensorRelaxedOneHotCategorical(temperature=torch.tensor(0.5))