Use observation as value_var of observed rvs

Author: ricardoV94

pymc/model.py

@@ -1275,7 +1275,7 @@ def make_obs_var(
 
             observed_rv_var.tag.observations = nonmissing_data
 
-            self.create_value_var(observed_rv_var, transform)
+            self.create_value_var(observed_rv_var, transform=None, value_var=nonmissing_data)
             self.add_random_variable(observed_rv_var, dims)
             self.observed_RVs.append(observed_rv_var)
 
@@ -1285,22 +1285,21 @@ def make_obs_var(
             rv_var = at.set_subtensor(rv_var[antimask_idx], observed_rv_var)
             rv_var = Deterministic(name, rv_var, self, dims, auto=True)
 
-        elif sps.issparse(data):
-            data = sparse.basic.as_sparse(data, name=name)
-            rv_var.tag.observations = data
-            self.create_value_var(rv_var, transform)
-            self.add_random_variable(rv_var, dims)
-            self.observed_RVs.append(rv_var)
         else:
-            data = at.as_tensor_variable(data, name=name)
+            if sps.issparse(data):
+                data = sparse.basic.as_sparse(data, name=name)
+            else:
+                data = at.as_tensor_variable(data, name=name)
             rv_var.tag.observations = data
-            self.create_value_var(rv_var, transform)
+            self.create_value_var(rv_var, transform=None, value_var=data)
             self.add_random_variable(rv_var, dims)
             self.observed_RVs.append(rv_var)
 
         return rv_var
 
-    def create_value_var(self, rv_var: TensorVariable, transform: Any) -> TensorVariable:
+    def create_value_var(
+        self, rv_var: TensorVariable, transform: Any, value_var: Optional[Variable] = None
+    ) -> TensorVariable:
         """Create a ``TensorVariable`` that will be used as the random
         variable's "value" in log-likelihood graphs.
 
@@ -1311,13 +1310,13 @@ def create_value_var(self, rv_var: TensorVariable, transform: Any) -> TensorVari
         this branch of the conditional.
 
         """
-        value_var = rv_var.type()
+        if value_var is None:
+            value_var = rv_var.type()
+            value_var.name = rv_var.name
 
         if aesara.config.compute_test_value != "off":
             value_var.tag.test_value = rv_var.tag.test_value
 
-        value_var.name = rv_var.name
-
         rv_var.tag.value_var = value_var
 
         # Make the value variable a transformed value variable,

pymc/tests/test_missing.py

@@ -19,8 +19,7 @@
 
 from numpy import array, ma
 
-from pymc.distributions.continuous import Gamma, Normal, Uniform
-from pymc.distributions.transforms import interval
+from pymc.distributions import Gamma, Normal, Uniform
 from pymc.exceptions import ImputationWarning
 from pymc.model import Model
 from pymc.sampling import sample, sample_posterior_predictive, sample_prior_predictive
@@ -94,10 +93,10 @@ def test_interval_missing_observations():
         with pytest.warns(ImputationWarning):
             theta2 = Normal("theta2", mu=theta1, observed=obs2, rng=rng)
 
-        assert "theta1_observed_interval__" in model.named_vars
+        assert "theta1_observed" in model.named_vars
         assert "theta1_missing_interval__" in model.named_vars
-        assert isinstance(
-            model.rvs_to_values[model.named_vars["theta1_observed"]].tag.transform, interval
+        assert not hasattr(
+            model.rvs_to_values[model.named_vars["theta1_observed"]].tag, "transform"
         )
 
         prior_trace = sample_prior_predictive(return_inferencedata=False)

pymc/tests/test_smc.py

@@ -409,12 +409,12 @@ def test_multiple_simulators(self):
         a_val = m.rvs_to_values[a]
         sim1_val = m.rvs_to_values[sim1]
         logp_sim1 = pm.logpt(sim1, sim1_val)
-        logp_sim1_fn = aesara.function([sim1_val, a_val], logp_sim1)
+        logp_sim1_fn = aesara.function([a_val], logp_sim1)
 
         b_val = m.rvs_to_values[b]
         sim2_val = m.rvs_to_values[sim2]
         logp_sim2 = pm.logpt(sim2, sim2_val)
-        logp_sim2_fn = aesara.function([sim2_val, b_val], logp_sim2)
+        logp_sim2_fn = aesara.function([b_val], logp_sim2)
 
         assert any(
             node for node in logp_sim1_fn.maker.fgraph.toposort() if isinstance(node.op, SortOp)