V0.2.1 (#74)

* Switch eig cals to eigh calls so we don't have issues with complex eigenvalues. Covariance is wrong right now anyways though I think

* fix the versioin of jax

* fix version of xspline

* change version to 0.2.1

---------

Co-authored-by: Alexander Hsu <owlx@uw.edu>

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "regmod"
-version = "0.2.0"
+version = "0.2.1"
 description = "General regression models"
 readme = "README.rst"
 requires-python = ">=3.10"
@@ -16,10 +16,10 @@ dependencies = [
     "numpy",
     "scipy",
     "pandas",
-    "xspline",
+    "xspline==0.0.7",
     "msca",
-    "jaxlib",
-    "jax[cpu]",
+    "jax[cpu]==0.4.5",
+    "jaxlib==0.4.4",
 ]
 
 [project.optional-dependencies]

src/regmod/models/model.py

@@ -127,7 +127,8 @@ def get_vcov(self, coefs: ndarray) -> ndarray:
         hessian = self.hessian(coefs)
         if isinstance(hessian, Matrix):
             hessian = hessian.to_numpy()
-        eig_vals, eig_vecs = np.linalg.eig(hessian)
+        #We probably don't want to be eigendecomposing
+        eig_vals, eig_vecs = np.linalg.eigh(hessian)
         if np.isclose(eig_vals, 0.0).any():
             raise ValueError("singular Hessian matrix, please add priors or "
                              "reduce number of variables")
@@ -136,7 +137,7 @@ def get_vcov(self, coefs: ndarray) -> ndarray:
         jacobian2 = self.jacobian2(coefs)
         if isinstance(jacobian2, Matrix):
             jacobian2 = jacobian2.to_numpy()
-        eig_vals = np.linalg.eigvals(jacobian2)
+        eig_vals = np.linalg.eigvalsh(jacobian2)
         if np.isclose(eig_vals, 0.0).any():
             raise ValueError("singular Jacobian matrix, please add priors or "
                              "reduce number of variables")