Reduce unnecessary iteration counts in slow tests

Targets tests identified by ricardoV94 in #7686:
- test_weakref_leak: reduce warmup+check from 20 to 8 iterations
- test_default_value_transform_logprob: reduce loop from 10 to 3
- test_interpolated (TestMatchesScipy): reduce x_points from 100k to 20k
- test_progressbar_nested_compound: reduce draws/tune from 10 to 5
- TestLKJCorr: reduce n values from (2,10,50) to (2,50)
- TestLKJCholeskyCov: reduce sizes_to_check from 7 to 4
- TestLKJCholeskCov: reduce parametrize combos from 6 to 4
- TestInterpolated: reduce mu*sigma loop from 56 to 4 combos

Not modified: test_mvnormal_indef (26.8s) — slowness is PyTensor
compilation overhead, not reducible without removing coverage.

Author: mcwanza

tests/distributions/test_continuous.py

@@ -965,7 +965,7 @@ class TestedInterpolated(pm.Interpolated):
 
                     @classmethod
                     def dist(cls, **kwargs):
-                        x_points = np.linspace(xmin, xmax, 100000)
+                        x_points = np.linspace(xmin, xmax, 20000)
                         pdf_points = st.norm.pdf(x_points, loc=mu, scale=sigma)
                         return super().dist(x_points=x_points, pdf_points=pdf_points, **kwargs)
 
@@ -2503,25 +2503,30 @@ def interpolated_rng_fn(self, size, mu, sigma, rng):
     ]
 
     def check_draws(self):
-        for mu in R.vals:
-            for sigma in Rplus.vals:
-                rng = self.get_random_state()
-
-                def ref_rand(size):
-                    return st.norm.rvs(loc=mu, scale=sigma, size=size, random_state=rng)
-
-                class TestedInterpolated(pm.Interpolated):
-                    rv_op = interpolated
-
-                    @classmethod
-                    def dist(cls, **kwargs):
-                        x_points = np.linspace(mu - 5 * sigma, mu + 5 * sigma, 100)
-                        pdf_points = st.norm.pdf(x_points, loc=mu, scale=sigma)
-                        return super().dist(x_points=x_points, pdf_points=pdf_points, **kwargs)
-
-                continuous_random_tester(
-                    TestedInterpolated,
-                    {},
-                    extra_args={"rng": pytensor.shared(rng)},
-                    ref_rand=ref_rand,
-                )
+        representative_params = [
+            (0.0, 1.0),
+            (-2.1, 0.1),
+            (1.0, 100.0),
+            (0.01, 0.99),
+        ]
+        for mu, sigma in representative_params:
+            rng = self.get_random_state()
+
+            def ref_rand(size):
+                return st.norm.rvs(loc=mu, scale=sigma, size=size, random_state=rng)
+
+            class TestedInterpolated(pm.Interpolated):
+                rv_op = interpolated
+
+                @classmethod
+                def dist(cls, **kwargs):
+                    x_points = np.linspace(mu - 5 * sigma, mu + 5 * sigma, 100)
+                    pdf_points = st.norm.pdf(x_points, loc=mu, scale=sigma)
+                    return super().dist(x_points=x_points, pdf_points=pdf_points, **kwargs)
+
+            continuous_random_tester(
+                TestedInterpolated,
+                {},
+                extra_args={"rng": pytensor.shared(rng)},
+                ref_rand=ref_rand,
+            )

tests/distributions/test_multivariate.py

@@ -980,7 +980,6 @@ def test_no_warning_logp(self):
         "size, shape",
         [
             ((10,), None),
-            (None, (10, 6)),
             (None, (10, ...)),
         ],
     )
@@ -2176,7 +2175,7 @@ def ref_rand(size, n, eta):
 
         # If passed as a domain, continuous_random_tester would make `n` a shared variable
         # But this RV needs it to be constant in order to define the inner graph
-        for n in (2, 10, 50):
+        for n in (2, 50):
             continuous_random_tester(
                 _LKJCorr,
                 {
@@ -2214,12 +2213,9 @@ class TestLKJCholeskyCov(BaseTestDistributionRandom):
     expected_rv_op_params = {"n": 3, "eta": 1.0, "sd_dist": pm.DiracDelta.dist([0.5, 1.0, 2.0])}
     size = None
 
-    sizes_to_check = [None, (), 1, (1,), 5, (4, 5), (2, 4, 2)]
+    sizes_to_check = [None, 5, (4, 5), (2, 4, 2)]
     sizes_expected = [
         (6,),
-        (6,),
-        (1, 6),
-        (1, 6),
         (5, 6),
         (4, 5, 6),
         (2, 4, 2, 6),

tests/logprob/test_transform_value.py

@@ -248,7 +248,7 @@ def test_default_value_transform_logprob(pt_dist, dist_params, sp_dist, size):
         on_unused_input="ignore",
     )
 
-    for i in range(10):
+    for i in range(3):
         a_dist = sp_dist(*dist_params)
         a_val = a_dist.rvs(size=size, random_state=test_val_rng).astype(a_value_var.dtype)
         b_dist = sp.stats.norm(a_val, 1.0)
@@ -614,9 +614,10 @@ def _growth(limit=10, peak_stats={}):
     rvs_to_values = {pt.random.beta(1, 1, name=f"p_{i}"): pt.scalar(f"p_{i}") for i in range(30)}
     tr = TransformValuesRewrite(dict.fromkeys(rvs_to_values.values(), logodds))
 
-    for i in range(20):
+    n_warmup = 4
+    n_check = 4
+    for i in range(n_warmup + n_check):
         conditional_logp(rvs_to_values, extra_rewrites=tr)
         res = _growth()
-        # Only start checking after warmup
-        if i > 15:
-            assert not res, "Object counts are still growing"
+        if i >= n_warmup:
+            assert not res, f"Object counts still growing after {n_warmup} warmup iterations: {res}"

tests/progress_bar/test_manager.py

@@ -35,8 +35,8 @@ def test_progressbar_nested_compound():
         )
 
         kwargs = {
-            "draws": 10,
-            "tune": 10,
+            "draws": 5,
+            "tune": 5,
             "chains": 2,
             "compute_convergence_checks": False,
             "step": step,