Fix scalar Gaussian log-pdf backend shape

src/pyrecest/distributions/nonperiodic/gaussian_distribution.py

@@ -181,6 +181,7 @@ def ln_pdf(self, xs):
         many likelihoods are accumulated or when densities may underflow.
         """
         xs = self._validate_evaluation_points(xs)
+        scalar_input = self.dim == 1 and ndim(xs) == 0
         if pyrecest.backend.__backend_name__ == "numpy":
             from scipy.stats import multivariate_normal as mvn
 
@@ -207,6 +208,8 @@ def ln_pdf(self, xs):
         else:
             raise NotImplementedError("Backend not supported")
 
+        if scalar_input:
+            log_pdf_vals = reshape(log_pdf_vals, ())
         return log_pdf_vals
 
     log_pdf = ln_pdf

tests/backend_support/test_gaussian_scalar_logpdf_contract.py

@@ -0,0 +1,36 @@
+import importlib.util
+
+import pytest
+
+from tests.support.backend_runner import run_backend_code
+
+pytestmark = pytest.mark.backend_portable
+
+
+@pytest.mark.parametrize("backend_name", ["numpy", "pytorch", "jax"])
+def test_one_dimensional_gaussian_scalar_logpdf_preserves_scalar_shape(backend_name):
+    if backend_name == "pytorch" and importlib.util.find_spec("torch") is None:
+        pytest.skip("PyTorch is not installed")
+    if backend_name == "jax" and importlib.util.find_spec("jax") is None:
+        pytest.skip("JAX is not installed")
+
+    code = """
+import pyrecest.backend as backend
+from pyrecest.distributions import GaussianDistribution
+
+
+distribution = GaussianDistribution(backend.array(0.0), backend.array(1.0))
+log_density = distribution.ln_pdf(backend.array(0.0))
+density = distribution.pdf(backend.array(0.0))
+batch_log_density = distribution.ln_pdf(backend.array([0.0]))
+
+assert tuple(backend.shape(backend.asarray(log_density))) == ()
+assert tuple(backend.shape(backend.asarray(density))) == ()
+assert tuple(backend.shape(backend.asarray(batch_log_density))) == (1,)
+assert float(backend.to_numpy(log_density)) < 0.0
+assert float(backend.to_numpy(density)) > 0.0
+print("ok")
+"""
+    result = run_backend_code(backend_name, code)
+    assert result.returncode == 0, result.stderr
+    assert "ok" in result.stdout