Add a negloglik feature

flip/fitter.py

@@ -163,7 +163,7 @@ def init_from_covariance(
         data,
         parameter_dict,
         likelihood_type="multivariate_gaussian",
-        likelihood_properties=None,
+        likelihood_properties={},
         **kwargs,
     ):
         """
@@ -194,7 +194,7 @@ def init_from_covariance(
         likelihood = minuit_fitter.get_likelihood(
             parameter_dict,
             likelihood_type=likelihood_type,
-            likelihood_properties=likelihood_properties,
+            likelihood_properties={**likelihood_properties, 'nloglik': True},
             **kwargs,
         )
         minuit_fitter.likelihood = likelihood
@@ -247,7 +247,7 @@ def init_from_file(
             data,
             parameter_dict,
             likelihood_type=likelihood_type,
-            likelihood_properties=likelihood_properties,
+            likelihood_properties={**likelihood_properties, 'nloglik': True},
         )
 
     def setup_minuit(self, parameter_dict):

flip/likelihood.py

@@ -11,49 +11,44 @@ def log_likelihood_gaussian_inverse(vector, covariance_sum):
     _, logdet = np.linalg.slogdet(covariance_sum)
     inverse_covariance_sum = np.linalg.inv(covariance_sum)
     chi2 = np.dot(vector, np.dot(inverse_covariance_sum, vector))
-    return 0.5 * (vector.size * np.log(2 * np.pi) + logdet + chi2)
+    return -0.5 * (vector.size * np.log(2 * np.pi) + logdet + chi2)
 
 
 def log_likelihood_gaussian_cholesky(vector, covariance_sum):
     cholesky = sc.linalg.cho_factor(covariance_sum)
     logdet = 2 * np.sum(np.log(np.diag(cholesky[0])))
     chi2 = np.dot(vector, sc.linalg.cho_solve(cholesky, vector))
-    return 0.5 * (vector.size * np.log(2 * np.pi) + logdet + chi2)
+    return -0.5 * (vector.size * np.log(2 * np.pi) + logdet + chi2)
 
 
 class BaseLikelihood(object):
+
+    _default_likelihood_properties = {
+            "inversion_method": "inverse",
+            "velocity_type": "direct",
+            "velocity_estimator": "full",
+        }
+
     def __init__(
         self,
         covariance=None,
         data=None,
         parameter_names=None,
-        likelihood_properties=None,
+        likelihood_properties={},
     ):
         self.covariance = covariance
         self.data = data
         self.parameter_names = parameter_names
 
-        _default_likelihood_properties = {
-            "inversion_method": "inverse",
-            "velocity_type": "direct",
-            "velocity_estimator": "full",
-        }
-        if likelihood_properties == None:
-            likelihood_properties = _default_likelihood_properties
-        else:
-            for key in _default_likelihood_properties.keys():
-                if key not in likelihood_properties.keys():
-                    likelihood_properties[key] = _default_likelihood_properties[key]
-
-        self.likelihood_properties = likelihood_properties
+        self.likelihood_properties =  {**self._default_likelihood_properties, **likelihood_properties}
 
     @classmethod
     def init_from_covariance(
         cls,
         covariance,
         data,
         parameter_names,
-        likelihood_properties=None,
+        likelihood_properties={},
         **kwargs,
     ):
         """
@@ -116,12 +111,18 @@ def load_data_vector(
 
 
 class MultivariateGaussianLikelihood(BaseLikelihood):
+    _default_likelihood_properties = {
+        'nloglik': False,
+        **BaseLikelihood._default_likelihood_properties
+        }
+
     def __init__(
         self,
         covariance=None,
         data=None,
         parameter_names=None,
-        likelihood_properties=None,
+        likelihood_properties={},
+        negloglik=False
     ):
         super(MultivariateGaussianLikelihood, self).__init__(
             covariance=covariance,
@@ -130,6 +131,7 @@ def __init__(
             likelihood_properties=likelihood_properties,
         )
 
+
     def verify_covariance(self):
         if self.covariance.full_matrix is False:
             self.covariance.compute_full_matrix()
@@ -147,6 +149,8 @@ def __call__(self, parameter_values):
         likelihood_function = eval(
             f"log_likelihood_gaussian_{self.likelihood_properties['inversion_method']}"
         )
+        if self.likelihood_properties['nloglik']:
+            return -likelihood_function(vector, covariance_sum)
         return likelihood_function(vector, covariance_sum)
 
 
@@ -156,7 +160,7 @@ def __init__(
         covariance=None,
         data=None,
         parameter_names=None,
-        likelihood_properties=None,
+        likelihood_properties={},
         interpolation_value_name=None,
         interpolation_value_range=None,
     ):
@@ -241,7 +245,7 @@ def __init__(
         covariance=None,
         data=None,
         parameter_names=None,
-        likelihood_properties=None,
+        likelihood_properties={},
         interpolation_value_range_0=None,
         interpolation_value_range_1=None,
     ):