docs: document distribution parameters and fix stale scalar rotation example

docs/reference/random_parameters.md

@@ -14,3 +14,27 @@ the specification directly and the transform samples from it at apply time:
 ## Choice
 
 ::: torchio.Choice
+
+## Distribution
+
+For continuous randomness, pass any `torch.distributions.Distribution`.
+The transform draws a fresh sample from it each time it is applied, so
+you can use any distribution supported by PyTorch:
+
+<!-- pytest-codeblocks:skip -->
+```python
+import torch
+import torchio as tio
+
+# Sample the rotation angle from a normal distribution (mean 0°, std 5°)
+transform = tio.Affine(degrees=torch.distributions.Normal(0, 5))
+```
+
+A distribution can also be combined per axis with fixed values and
+ranges:
+
+<!-- pytest-codeblocks:skip -->
+```python
+# Fixed along I, uniform range along J, normal distribution along K
+transform = tio.Affine(degrees=(0, (-10, 10), torch.distributions.Normal(0, 5)))
+```

docs/tutorials/augmentation.md

@@ -41,7 +41,7 @@ a tuple gives a uniform range:
 tio.Affine(degrees=90)
 
 # Rotate uniformly between -15° and 15°
-tio.Affine(degrees=15)
+tio.Affine(degrees=(-15, 15))
 
 # Rotate uniformly between 5° and 20°
 tio.Affine(degrees=(5, 20))
@@ -63,6 +63,18 @@ You can mix `Choice`, ranges, and fixed values per axis:
 tio.Affine(degrees=(0, (-10, 10), tio.Choice([-90, 0, 90])))
 ```
 
+For continuous sampling beyond a uniform range, pass any
+`torch.distributions.Distribution`.  A fresh value is drawn each time
+the transform is applied:
+
+<!-- pytest-codeblocks:skip -->
+```python
+import torch
+
+# Sample the rotation angle from a normal distribution (mean 0°, std 5°)
+tio.Affine(degrees=torch.distributions.Normal(0, 5))
+```
+
 ## Probability control
 
 Every transform has a `p` parameter (probability of being applied):