0.26.10 det fixed, new plots

Author: bartzbeielstein

notebooks/00_spotPython_tests.ipynb

@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "spotpython"
-version = "0.26.9"
+version = "0.26.10"
 authors = [
   { name="T. Bartz-Beielstein", email="tbb@bartzundbartz.de" }
 ]

pyproject.toml

@@ -667,7 +667,10 @@ def _build(self) -> None:
         #     raise RuntimeError("Covariance matrix has already been built.")
         self.K = covar_anisotropic(self.X, d=self.d, g=self.g)
         self.Ki = matrix_inversion_dispatcher(self.K, method=self.nlsep_method)
-        self.ldetK = np.log(det(self.K))
+        detK = det(self.K)
+        if detK <= 1e-14:
+            detK = 1e-14  # TODO: Check if this can be improved
+        self.ldetK = np.log(detK)
         self.calc_ytKiy()
         # TODO: Check if this is necessary
         # if self.dK:

src/spotpython/gp/gp_sep.py

@@ -9,10 +9,12 @@ def nlsep(par, X, Y, nlsep_method="inv") -> float:
     Calculate the negative log-likelihood for a separable power exponential correlation function.
 
     Args:
-        par (np.ndarray): Array of parameters, where the first ncol(X) elements are the range parameters
-                          and the last element is the nugget parameter.
-        X (np.ndarray): Input matrix of shape (n, col).
-        Y (np.ndarray): Response vector of shape (n,).
+        par (np.ndarray):
+            Array of parameters, where the first ncol(X) elements are the range parameters and the last element is the nugget parameter.
+        X (np.ndarray):
+            Input matrix of shape (n, col).
+        Y (np.ndarray):
+            Response vector of shape (n,).
 
     Returns:
         float: Negative log-likelihood.
@@ -32,7 +34,10 @@ def nlsep(par, X, Y, nlsep_method="inv") -> float:
     n = len(Y)
     K = covar_anisotropic(X, d=theta, g=g)
     Ki = matrix_inversion_dispatcher(K, method=nlsep_method)
-    ldetK = np.log(det(K))
+    detK = det(K)
+    if detK <= 1e-14:
+        detK = 1e-14  # TODO: Check if this can be improved
+    ldetK = np.log(detK)
     ll = -(n / 2) * np.log(Y.T @ Ki @ Y) - (1 / 2) * ldetK
     return -ll