Fix join logp for multivariate RVs

pymc/logprob/tensor.py

@@ -187,10 +187,12 @@ def logprob_join(op, values, axis, *base_rvs, **kwargs):
     # If the stacked variables depend on each other, we have to replace them by the respective values
     logps = replace_rvs_by_values(logps, rvs_to_values=base_rvs_to_split_values)
 
-    base_vars_ndim_supp = split_values[0].ndim - logps[0].ndim
+    # Make axis positive and adjust for multivariate logp fewer dimensions to the right
+    axis = pt.switch(axis >= 0, axis, value.ndim + axis)
+    axis = pt.minimum(axis, logps[0].ndim - 1)
     join_logprob = pt.concatenate(
         [pt.atleast_1d(logp) for logp in logps],
-        axis=axis - base_vars_ndim_supp,
+        axis=axis,
     )
 
     return join_logprob

tests/logprob/test_tensor.py

@@ -269,34 +269,23 @@ def test_measurable_join_univariate(size1, size2, axis, concatenate):
 
 
 @pytest.mark.parametrize(
-    "size1, supp_size1, size2, supp_size2, axis, concatenate",
+    "size1, supp_size1, size2, supp_size2, axis, concatenate, logp_axis",
     [
-        (None, 2, None, 2, 0, True),
-        (None, 2, None, 2, -1, True),
-        ((5,), 2, (3,), 2, 0, True),
-        ((5,), 2, (3,), 2, -2, True),
-        ((2,), 5, (2,), 3, 1, True),
-        pytest.param(
-            (2,),
-            5,
-            (2,),
-            5,
-            0,
-            False,
-            marks=pytest.mark.xfail(reason="cannot measure dimshuffled multivariate RVs"),
-        ),
-        pytest.param(
-            (2,),
-            5,
-            (2,),
-            5,
-            1,
-            False,
-            marks=pytest.mark.xfail(reason="cannot measure dimshuffled multivariate RVs"),
-        ),
+        (None, 2, None, 2, 0, True, 0),
+        (None, 2, None, 2, -1, True, 0),
+        ((5,), 2, (3,), 2, 0, True, 0),
+        ((5,), 2, (3,), 2, -2, True, 0),
+        ((2,), 5, (2,), 3, 1, True, 0),
+        ((5, 6), 10, (5, 1), 10, 1, True, 1),
+        ((5, 6), 10, (5, 1), 10, -2, True, 1),
+        ((2,), 5, (2,), 5, 0, False, 0),
+        ((2,), 5, (2,), 5, 1, False, 1),
+        ((5, 6), 10, (5, 6), 10, 2, False, 2),
     ],
 )
-def test_measurable_join_multivariate(size1, supp_size1, size2, supp_size2, axis, concatenate):
+def test_measurable_join_multivariate(
+    size1, supp_size1, size2, supp_size2, axis, concatenate, logp_axis
+):
     base1_rv = pt.random.multivariate_normal(
         np.zeros(supp_size1), np.eye(supp_size1), size=size1, name="base1"
     )
@@ -310,19 +299,18 @@ def test_measurable_join_multivariate(size1, supp_size1, size2, supp_size2, axis
     base1_vv = base1_rv.clone()
     base2_vv = base2_rv.clone()
     y_vv = y_rv.clone()
+
+    y_logp = logp(y_rv, y_vv)
+    assert_no_rvs(y_logp)
+
     base_logps = [
         pt.atleast_1d(logp)
         for logp in conditional_logp({base1_rv: base1_vv, base2_rv: base2_vv}).values()
     ]
-
     if concatenate:
-        axis_norm = np.core.numeric.normalize_axis_index(axis, base1_rv.ndim)
-        base_logps = pt.concatenate(base_logps, axis=axis_norm - 1)
+        expected_logp = pt.concatenate(base_logps, axis=logp_axis)
     else:
-        axis_norm = np.core.numeric.normalize_axis_index(axis, base1_rv.ndim + 1)
-        base_logps = pt.stack(base_logps, axis=axis_norm - 1)
-    y_logp = y_logp = logp(y_rv, y_vv)
-    assert_no_rvs(y_logp)
+        expected_logp = pt.stack(base_logps, axis=logp_axis)
 
     base1_testval = base1_rv.eval()
     base2_testval = base2_rv.eval()
@@ -331,7 +319,7 @@ def test_measurable_join_multivariate(size1, supp_size1, size2, supp_size2, axis
     else:
         y_testval = np.stack((base1_testval, base2_testval), axis=axis)
     np.testing.assert_allclose(
-        base_logps.eval({base1_vv: base1_testval, base2_vv: base2_testval}),
+        expected_logp.eval({base1_vv: base1_testval, base2_vv: base2_testval}),
         y_logp.eval({y_vv: y_testval}),
     )