ENH allow all kinds of covariances (#104)

* ENH allow all kinds of covariances

* TST remove outdated tests

Author: beckermr

firecrown/ccl/likelihoods/gaussian.py

@@ -1,6 +1,5 @@
 import numpy as np
 import scipy.linalg
-import sacc
 
 from ..core import LogLike
 
@@ -42,10 +41,6 @@ def read(self, sacc_data, sources, statistics):
             A dictionary mapping statistics to their objects. These statistics do
             not have to have been rendered.
         """
-        if not isinstance(sacc_data.covariance, sacc.covariance.FullCovariance):
-            raise RuntimeError(
-                "Currently, the SACC covariance must be a 'FullCovariance'. "
-                "This may change in a future firecrown release.")
         _sd = sacc_data.copy()
         inds = []
         for stat in self.data_vector:
@@ -58,7 +53,7 @@ def read(self, sacc_data, sources, statistics):
         cov = np.zeros((len(inds), len(inds)))
         for new_i, old_i in enumerate(inds):
             for new_j, old_j in enumerate(inds):
-                cov[new_i, new_j] = _sd.covariance.covmat[old_i, old_j]
+                cov[new_i, new_j] = _sd.covariance.dense[old_i, old_j]
         self.cov = cov
         self.cholesky = scipy.linalg.cholesky(self.cov, lower=True)
 

firecrown/ccl/likelihoods/tdist.py

@@ -1,6 +1,5 @@
 import numpy as np
 import scipy.linalg
-import sacc
 
 from ..core import LogLike
 
@@ -50,10 +49,6 @@ def read(self, sacc_data, sources, statistics):
             A dictionary mapping statistics to their objects. These statistics do
             not have to have been rendered.
         """
-        if not isinstance(sacc_data.covariance, sacc.covariance.FullCovariance):
-            raise RuntimeError(
-                "Currently, the SACC covariance must be a 'FullCovariance'. "
-                "This may change in a future firecrown release.")
         _sd = sacc_data.copy()
         inds = []
         for stat in self.data_vector:
@@ -66,7 +61,7 @@ def read(self, sacc_data, sources, statistics):
         cov = np.zeros((len(inds), len(inds)))
         for new_i, old_i in enumerate(inds):
             for new_j, old_j in enumerate(inds):
-                cov[new_i, new_j] = _sd.covariance.covmat[old_i, old_j]
+                cov[new_i, new_j] = _sd.covariance.dense[old_i, old_j]
         self.cov = cov
         self.cholesky = scipy.linalg.cholesky(self.cov, lower=True)
 

firecrown/ccl/likelihoods/tests/test_gaussian.py

@@ -1,5 +1,4 @@
 import numpy as np
-import pytest
 
 from ..gaussian import ConstGaussianLogLike
 
@@ -34,15 +33,3 @@ def test_likelihood_gaussian_subset(likelihood_test_data):
     cov = likelihood_test_data['cov'][0:4, 0:4]
     loglike = -0.5 * np.dot(delta, np.dot(np.linalg.inv(cov), delta))
     assert np.allclose(loglike, ll.compute(data, theory))
-
-
-def test_likelihood_gaussian_raises(likelihood_test_data):
-    ll = ConstGaussianLogLike(data_vector=["stat_src0_src0", "stat_src0_src1"])
-    sd = likelihood_test_data['sacc_data'].copy()
-    sd.covariance = 10
-    with pytest.raises(RuntimeError) as e:
-        ll.read(
-            sd,
-            likelihood_test_data['sources'],
-            likelihood_test_data['statistics'])
-    assert 'FullCovariance' in str(e)

firecrown/ccl/likelihoods/tests/test_tdist.py

@@ -1,5 +1,4 @@
 import numpy as np
-import pytest
 
 from ..tdist import TdistLogLike
 
@@ -42,15 +41,3 @@ def test_likelihood_tdist_subset(likelihood_test_data):
     chi2 = np.dot(delta, np.dot(np.linalg.inv(cov), delta))
     loglike = -0.5 * nu * np.log(1.0 + chi2 / (nu - 1.0))
     assert np.allclose(loglike, ll.compute(data, theory))
-
-
-def test_likelihood_tdist_raises(likelihood_test_data):
-    ll = TdistLogLike(data_vector=["stat_src0_src0", "stat_src0_src1"], nu=10)
-    sd = likelihood_test_data['sacc_data'].copy()
-    sd.covariance = 10
-    with pytest.raises(RuntimeError) as e:
-        ll.read(
-            sd,
-            likelihood_test_data['sources'],
-            likelihood_test_data['statistics'])
-    assert 'FullCovariance' in str(e)