Merge branch 'cross-cov'

Author: andreicuceu

vega/build_config.py

@@ -392,6 +392,8 @@ def _build_main_config(self, fit_type, fit_info, parameters, git_hash):
         config['data sets']['zeff'] = str(self.zeff_in)
         corr_paths = [str(path) for path in self.corr_paths]
         config['data sets']['ini files'] = ' '.join(corr_paths)
+        if 'global_cov_file' in fit_info:
+            config['data sets']['global-cov-file'] = fit_info.get('global_cov_file')
 
         # Write the scale parameters functions
         config['cosmo-fit type'] = {}

vega/data.py

@@ -4,7 +4,7 @@
 from scipy import sparse
 from scipy.sparse import csr_matrix
 
-from vega.utils import find_file
+from vega.utils import find_file, compute_masked_invcov, compute_log_cov_det
 from vega.coordinates import Coordinates
 
 BLINDING_STRATEGIES = ['desi_m2', 'desi_y1']
@@ -159,7 +159,8 @@ def inv_masked_cov(self):
         """
         if self._inv_masked_cov is None:
             # Compute inverse of the covariance matrix
-            self.compute_masked_invcov()
+            self._inv_masked_cov = compute_masked_invcov(
+                self.cov_mat, self.data_mask, self.invert_full_cov)
 
         return self._inv_masked_cov
 
@@ -173,13 +174,7 @@ def log_cov_det(self):
             Logarithm of the determinant of the covariance matrix
         """
         if self._log_cov_det is None:
-            # Compute the log determinant using and LDL^T decomposition
-            # |C| = Product of Diagonal components of D
-            masked_cov = self.cov_mat[:, self.data_mask]
-            masked_cov = masked_cov[self.data_mask, :]
-            _, d, __ = linalg.ldl(masked_cov)
-            self._log_cov_det = np.log(d.diagonal()).sum()
-            assert isinstance(self.log_cov_det, float)
+            self._log_cov_det = compute_log_cov_det(self.cov_mat, self.data_mask)
 
         return self._log_cov_det
 
@@ -619,32 +614,3 @@ def create_monte_carlo(self, fiducial_model, scale=None, seed=None, forecast=Fal
         self.masked_mc_mock = self.mc_mock[self.data_mask]
 
         return self.mc_mock
-
-    def compute_masked_invcov(self):
-        """Compute the masked inverse of the covariance matrix.
-        """
-        masked_cov = self.cov_mat[:, self.data_mask]
-        masked_cov = masked_cov[self.data_mask, :]
-
-        try:
-            linalg.cholesky(self.cov_mat)
-            print('LOG: Full matrix is positive definite')
-        except linalg.LinAlgError:
-            if self.invert_full_cov:
-                raise ValueError('Full matrix is not positive definite. '
-                                 'Use invert-full-cov = False to work with the masked covariance')
-            else:
-                print('WARNING: Full matrix is not positive definite')
-
-        try:
-            linalg.cholesky(masked_cov)
-            print('LOG: Reduced matrix is positive definite')
-        except linalg.LinAlgError:
-            print('WARNING: Reduced matrix is not positive definite')
-
-        if self.invert_full_cov:
-            inv_cov = linalg.inv(self.cov_mat)
-            self._inv_masked_cov = inv_cov[:, self.data_mask]
-            self._inv_masked_cov = self._inv_masked_cov[self.data_mask, :]
-        else:
-            self._inv_masked_cov = linalg.inv(masked_cov)

vega/utils.py

@@ -228,5 +228,66 @@ def find_file(path):
     raise RuntimeError('The path/file does not exists: ', input_path)
 
 
+def compute_masked_invcov(cov_mat, data_mask, invert_full_cov=False):
+    """Compute the masked inverse of the covariance matrix
+
+    Parameters
+    ----------
+    cov_mat : Array
+        Covariance matrix
+    data_mask : Array
+        Mask of the data
+    invert_full_cov : bool, optional
+        Flag to invert the full covariance matrix, by default False
+    """
+    masked_cov = cov_mat[:, data_mask]
+    masked_cov = masked_cov[data_mask, :]
+
+    try:
+        np.linalg.cholesky(cov_mat)
+        print('LOG: Full matrix is positive definite')
+    except np.linalg.LinAlgError:
+        if invert_full_cov:
+            raise ValueError('Full matrix is not positive definite. '
+                             'Use invert-full-cov = False to work with the masked covariance')
+        else:
+            print('WARNING: Full matrix is not positive definite')
+
+    try:
+        np.linalg.cholesky(masked_cov)
+        print('LOG: Reduced matrix is positive definite')
+    except np.linalg.LinAlgError:
+        print('WARNING: Reduced matrix is not positive definite')
+
+    if invert_full_cov:
+        inv_cov = np.linalg.inv(cov_mat)
+        inv_masked_cov = inv_cov[:, data_mask]
+        inv_masked_cov = inv_masked_cov[data_mask, :]
+    else:
+        inv_masked_cov = np.linalg.inv(masked_cov)
+
+    return inv_masked_cov
+
+
+def compute_log_cov_det(cov_mat, data_mask):
+    """Compute the log of the determinant of the covariance matrix
+
+    Parameters
+    ----------
+    cov_mat : Array
+        Covariance matrix
+    data_mask : Array
+        Mask of the data
+
+    Returns
+    -------
+    float
+        Log of the determinant of the covariance matrix
+    """
+    masked_cov = cov_mat[:, data_mask]
+    masked_cov = masked_cov[data_mask, :]
+    return np.linalg.slogdet(masked_cov)[1]
+
+
 class VegaBoundsError(Exception):
     pass