Rename _replace_rvs_in_graphs and fix bug when replacing input

pymc/logprob/utils.py

@@ -337,7 +337,7 @@ def ignore_logprob_multiple_vars(
     making each "unmeasurable", whereas a sequential call to `ignore_logprob`
     would not do this correctly.
     """
-    from pymc.pytensorf import _replace_rvs_in_graphs
+    from pymc.pytensorf import _replace_vars_in_graphs
 
     measurable_vars_to_unmeasurable_vars = {
         measurable_var: ignore_logprob(measurable_var) for measurable_var in vars
@@ -353,5 +353,5 @@ def replacement_fn(var, replacements):
 
         return []
 
-    unmeasurable_vars, _ = _replace_rvs_in_graphs(graphs=vars, replacement_fn=replacement_fn)
+    unmeasurable_vars, _ = _replace_vars_in_graphs(graphs=vars, replacement_fn=replacement_fn)
     return unmeasurable_vars

pymc/pytensorf.py

@@ -205,19 +205,22 @@ def expand(var):
     yield from walk(graphs, expand, bfs=False)
 
 
-def _replace_rvs_in_graphs(
+def _replace_vars_in_graphs(
     graphs: Iterable[TensorVariable],
     replacement_fn: Callable[[TensorVariable], Dict[TensorVariable, TensorVariable]],
     **kwargs,
 ) -> Tuple[List[TensorVariable], Dict[TensorVariable, TensorVariable]]:
-    """Replace random variables in graphs
+    """Replace variables in graphs.
 
     This will *not* recompute test values.
 
     Parameters
     ----------
     graphs
         The graphs in which random variables are to be replaced.
+    replacement_fn
+        A callable called on each graph output that populates a replacement dictionary and returns
+        nodes that should be investigated further.
 
     Returns
     -------
@@ -256,7 +259,8 @@ def expand_replace(var):
         toposort = fg.toposort()
         sorted_replacements = sorted(
             tuple(replacements.items()),
-            key=lambda pair: toposort.index(pair[0].owner),
+            # Root inputs don't have owner, we give them negative priority -1
+            key=lambda pair: toposort.index(pair[0].owner) if pair[0].owner is not None else -1,
             reverse=True,
         )
         fg.replace_all(sorted_replacements, import_missing=True)
@@ -317,7 +321,7 @@ def populate_replacements(
     equiv = clone_get_equiv(inputs, graphs, False, False, {})
     graphs = [equiv[n] for n in graphs]
 
-    graphs, _ = _replace_rvs_in_graphs(
+    graphs, _ = _replace_vars_in_graphs(
         graphs,
         replacement_fn=populate_replacements,
         **kwargs,
@@ -385,7 +389,7 @@ def poulate_replacements(rv, replacements):
         # replacements if that is not a simple input variable
         return [value]
 
-    graphs, _ = _replace_rvs_in_graphs(
+    graphs, _ = _replace_vars_in_graphs(
         graphs,
         replacement_fn=poulate_replacements,
         **kwargs,

tests/test_pytensorf.py

@@ -24,6 +24,7 @@
 import pytest
 import scipy.sparse as sps
 
+from pytensor import shared
 from pytensor.compile.builders import OpFromGraph
 from pytensor.graph.basic import Variable, equal_computations
 from pytensor.tensor.random.basic import normal, uniform
@@ -40,6 +41,7 @@
 from pymc.exceptions import NotConstantValueError
 from pymc.logprob.utils import ParameterValueError
 from pymc.pytensorf import (
+    _replace_vars_in_graphs,
     collect_default_updates,
     compile_pymc,
     constant_fold,
@@ -821,3 +823,21 @@ def test_interdependent_transformed_rvs(self, reversed, test_deprecated_fn):
             ),
             [expected_x, expected_y, expected_z, expected_w],
         )
+
+    def test_replace_input(self):
+        inp = shared(0.0, name="inp")
+        x = pm.Normal.dist(inp)
+
+        assert x.eval() < 50
+
+        new_inp = inp + 100
+
+        def replacement_fn(var, replacements):
+            if var is x:
+                replacements[x.owner.inputs[3]] = new_inp
+
+            return []
+
+        [new_x], _ = _replace_vars_in_graphs([x], replacement_fn=replacement_fn)
+
+        assert new_x.eval() > 50