refactor(tensor_distribution): simplify TensorLaplace initialization

Leverage `torch.distributions.utils.broadcast_all` to streamline parameter broadcasting and type conversion, which also simplifies shape and device inference.

This change removes manual validation and broadcasting logic, making the implementation more concise and consistent with other distributions in the module. Redundant overrides for `log_prob` and `variance` have also been removed as they are handled by the base class.

Author: Tim Joseph

src/tensorcontainer/tensor_distribution/laplace.py

@@ -2,9 +2,9 @@
 
 from typing import Any, Dict, Optional
 
-import torch
 from torch import Tensor
 from torch.distributions import Laplace
+from torch.distributions.utils import broadcast_all
 
 from .base import TensorDistribution
 
@@ -22,21 +22,9 @@ def __init__(
         scale: Tensor | float,
         validate_args: Optional[bool] = None,
     ):
-        # Store the parameters in annotated attributes before calling super().__init__()
-        # This is required because super().__init__() calls self.dist() which needs these attributes
-        self._loc = loc if isinstance(loc, Tensor) else torch.tensor(loc)
-        self._scale = scale if isinstance(scale, Tensor) else torch.tensor(scale)
-
-        if torch.any(self._scale <= 0):
-            raise ValueError("scale must be positive")
-
-        try:
-            torch.broadcast_tensors(self._loc, self._scale)
-        except RuntimeError as e:
-            raise ValueError(f"loc and scale must have compatible shapes: {e}")
-
-        shape = torch.broadcast_shapes(self._loc.shape, self._scale.shape)
-        device = self._loc.device if self._loc.numel() > 0 else self._scale.device
+        self._loc, self._scale = broadcast_all(loc, scale)
+        shape = self._loc.shape
+        device = self._loc.device
         super().__init__(shape, device, validate_args)
 
     @classmethod
@@ -55,9 +43,6 @@ def dist(self) -> Laplace:
             validate_args=self._validate_args,
         )
 
-    def log_prob(self, value: Tensor) -> Tensor:
-        return self.dist().log_prob(value)
-
     @property
     def loc(self) -> Optional[Tensor]:
         """Returns the loc used to initialize the distribution."""
@@ -68,8 +53,4 @@ def scale(self) -> Optional[Tensor]:
         """Returns the scale used to initialize the distribution."""
         return self.dist().scale
 
-    @property
-    def variance(self) -> Tensor:
-        """Returns the variance of the Laplace distribution."""
-        assert self._scale is not None
-        return self.dist().variance
+