Fix assertion failures escaping Fitness.call

autofit/non_linear/fitness.py

@@ -168,17 +168,18 @@ def call(self, parameters):
         -------
         The figure of merit returned to the non-linear search, which is either the log likelihood or log posterior.
         """
-        # Get instance from model
-        instance = self.model.instance_from_vector(vector=parameters, xp=self._xp)
-
         if self._xp.__name__.startswith("jax"):
 
+            # Get instance from model (must be side-effect free and exception-free under JAX)
+            instance = self.model.instance_from_vector(vector=parameters, xp=self._xp)
+
             # Evaluate log likelihood (must be side-effect free and exception-free)
             log_likelihood = self.analysis.log_likelihood_function(instance=instance)
 
         else:
 
             try:
+                instance = self.model.instance_from_vector(vector=parameters, xp=self._xp)
                 log_likelihood = self.analysis.log_likelihood_function(instance=instance)
             except exc.FitException:
                 return self.resample_figure_of_merit

test_autofit/non_linear/test_fitness_assertions.py

@@ -0,0 +1,32 @@
+import numpy as np
+
+import autofit as af
+from autofit.non_linear.fitness import Fitness
+
+
+def test_fitness_returns_resample_fom_on_assertion_failure():
+    """
+    If an added assertion rejects a sampled parameter vector, ``Fitness.call``
+    must return ``resample_figure_of_merit`` rather than letting the
+    ``FitException`` escape into the non-linear search. Regression for
+    the DynestyStatic-blocking bug where ``instance_from_vector`` was
+    called outside the ``try/except FitException`` block.
+    """
+    gaussian_0 = af.Model(af.ex.Gaussian)
+    gaussian_1 = af.Model(af.ex.Gaussian)
+    gaussian_0.add_assertion(gaussian_0.centre > gaussian_1.centre)
+    model = af.Collection(gaussian_0=gaussian_0, gaussian_1=gaussian_1)
+
+    data = np.ones(20)
+    noise_map = np.ones(20) * 0.1
+    analysis = af.ex.Analysis(data=data, noise_map=noise_map)
+
+    fitness = Fitness(model=model, analysis=analysis)
+
+    # centre_0 (index 0) = 10 < centre_1 (index 3) = 20  → assertion violated
+    violating = [10.0, 1.0, 1.0, 20.0, 1.0, 1.0]
+    assert fitness.call(violating) == fitness.resample_figure_of_merit
+
+    # centre_0 = 20 > centre_1 = 10  → assertion satisfied, real FOM returned
+    satisfying = [20.0, 1.0, 1.0, 10.0, 1.0, 1.0]
+    assert fitness.call(satisfying) != fitness.resample_figure_of_merit