Add warning if a Minibatched variable is used but total_size is forgo…

…tten

This should help catch a well-known Minibatch footgun

pymc/model/core.py

@@ -40,7 +40,7 @@
 from pytensor.tensor.variable import TensorConstant, TensorVariable
 
 from pymc.blocking import DictToArrayBijection, RaveledVars
-from pymc.data import is_valid_observed
+from pymc.data import is_valid_observed, MinibatchOp
 from pymc.exceptions import (
     BlockModelAccessError,
     ImputationWarning,
@@ -1241,6 +1241,14 @@ def register_rv(
             self.add_named_variable(rv_var, dims)
             self.set_initval(rv_var, initval)
         else:
+            if (
+                isinstance(observed, TensorVariable)
+                and isinstance(observed.owner.op, MinibatchOp)
+                and total_size is None
+            ):
+                warnings.warn(
+                    "total_size not provided for observed variable that uses pm.Minibatch"
+                )
             if not is_valid_observed(observed):
                 raise TypeError(
                     "Variables that depend on other nodes cannot be used for observed data."

tests/variational/test_minibatch_rv.py

@@ -112,6 +112,13 @@ def test_random(self):
         assert mx is not x
         np.testing.assert_array_equal(draw(mx, random_seed=1), draw(x, random_seed=1))
 
+    def test_warning_on_missing_total_size(self):
+        total_size = 1000
+        with pytest.warns(match="total_size not provided"):
+            with pm.Model() as m:
+                MB = pm.Minibatch(np.arange(total_size, dtype="float64"), batch_size=100)
+                pm.Normal("n", observed=MB)
+
     @pytest.mark.filterwarnings("error")
     def test_minibatch_parameter_and_value(self):
         rng = np.random.default_rng(161)