Merge pull request #347 from johannbrehmer/develop

Fixes for SCANDAL (and Python 2)

madminer/fisherinformation.py

@@ -933,9 +933,7 @@ def histogram_of_fisher_information(
 
                 # Rescale for test_split
                 if test_split is not None:
-                    correction = np.array(
-                        [1./test_split for obs_event in observations]
-                    )
+                    correction = np.array([1.0 / test_split for obs_event in observations])
                     weights_benchmarks *= correction[:, np.newaxis]
 
                 weights_theta = mdot(theta_matrix, weights_benchmarks)
@@ -954,7 +952,7 @@ def histogram_of_fisher_information(
                     fisher_info_events = model.calculate_fisher_information(
                         x=observations,
                         weights=weights_theta,
-                         n_events=luminosity * np.sum(weights_theta) ,
+                        n_events=luminosity * np.sum(weights_theta),
                         sum_events=False,
                     )
 

madminer/limits.py

@@ -158,7 +158,7 @@ def _analyse(
             model = load_estimator(model_file)
 
             logger.info("Calculating kinematic log likelihood ratio with estimator")
-            log_r_kin = self._calculate_log_likelihood_ratio_kinematics(x, theta_grid, model,theta_true)
+            log_r_kin = self._calculate_log_likelihood_ratio_kinematics(x, theta_grid, model, theta_true)
             log_r_kin = log_r_kin.astype(np.float64)
             log_r_kin = self._clean_nans(log_r_kin)
             logger.debug("Raw mean -2 log r: %s", np.mean(-2.0 * log_r_kin, axis=1))
@@ -368,11 +368,8 @@ def _calculate_log_likelihood_ratio_kinematics(self, x_observed, theta_grid, mod
                 x=x_observed, theta=theta_grid, test_all_combinations=True, evaluate_score=False
             )
         elif isinstance(model, LikelihoodEstimator):
-            raise NotImplementedError(
-                "LikelihoodEstimator is not implemented!"
-            )
-            log_r, _ = model.evaluate_log_likelihood_ratio(
-                x=x_observed, theta0=theta_grid, theta1=np.array(theta1), test_all_combinations=True, evaluate_score=False
+            log_r, _ = model.evaluate_log_likelihood(
+                x=x_observed, theta=theta_grid, test_all_combinations=True, evaluate_score=False
             )
         elif isinstance(model, Ensemble) and model.estimator_type == "parameterized_ratio":
             log_r, _ = model.evaluate_log_likelihood_ratio(
@@ -382,10 +379,18 @@ def _calculate_log_likelihood_ratio_kinematics(self, x_observed, theta_grid, mod
                 evaluate_score=False,
                 calculate_covariance=False,
             )
+        elif isinstance(model, Ensemble) and model.estimator_type == "likelihood":
+            log_r, _ = model.evaluate_log_likelihood(
+                x=x_observed,
+                theta=theta_grid,
+                test_all_combinations=True,
+                evaluate_score=False,
+                calculate_covariance=False,
+            )
         else:
             raise NotImplementedError(
                 "Likelihood ratio estimation is currently only implemented for "
-                "ParameterizedRatioEstimator instancees"
+                "ParameterizedRatioEstimator and LikelihoodEstimator instancees"
             )
         return log_r
 

madminer/plotting.py

@@ -991,10 +991,10 @@ def plot_fisher_information_contours_2d(
         logger.debug("Std: %s", uncertainties)
 
     # Plot results
-    do_fig=False
+    do_fig = False
     if ax is None:
-        do_fig=True
-        fig,ax = plt.figure(figsize=(5.0, 5.0))
+        do_fig = True
+        fig, ax = plt.figure(figsize=(5.0, 5.0))
 
     # Error bands
     for i in range(n_matrices):
@@ -1036,6 +1036,7 @@ def plot_fisher_information_contours_2d(
     else:
         return ax
 
+
 def plot_fisherinfo_barplot(
     fisher_information_matrices, labels, determinant_indices=None, eigenvalue_colors=None, bar_colors=None
 ):
@@ -1246,10 +1247,10 @@ def plot_distribution_of_information(
         Plot as Matplotlib Figure instance.
 
     """
-    #prepare Plot
+    # prepare Plot
     if fontsize is not None:
-        matplotlib.rcParams.update({'font.size': fontsize})
-    
+        matplotlib.rcParams.update({"font.size": fontsize})
+
     # Prepare data
     n_entries = len(fisher_information_matrices)
     size = len(fisher_information_matrices[1])
@@ -1292,7 +1293,7 @@ def plot_distribution_of_information(
     # xsec plot
     fig = plt.figure(figsize=figsize)
     ax1 = plt.subplot(111)
-    #fig.subplots_adjust(left=0.1667, right=0.8333, bottom=0.17, top=0.97)
+    # fig.subplots_adjust(left=0.1667, right=0.8333, bottom=0.17, top=0.97)
 
     if log_xsec:
         ax1.set_yscale("log")

madminer/utils/ml/models/batch_norm.py

@@ -74,7 +74,7 @@ def generate_samples(self, n_samples=1, u=None, **kwargs):
     def to(self, *args, **kwargs):
         logger.debug("Transforming BatchNorm to %s", args)
 
-        self = super().to(*args, **kwargs)
+        self = super(BatchNorm, self).to(*args, **kwargs)
 
         self.running_mean = self.running_mean.to(*args, **kwargs)
         self.running_var = self.running_var.to(*args, **kwargs)

madminer/utils/ml/models/made.py

@@ -139,7 +139,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(GaussianMADE, self).to(*args, **kwargs)
 
         for i, (M, W, b) in enumerate(zip(self.Ms, self.Ws, self.bs)):
             self.Ms[i] = M.to(*args, **kwargs)
@@ -314,7 +314,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(ConditionalGaussianMADE, self).to(*args, **kwargs)
 
         for i, (M, W, b) in enumerate(zip(self.Ms, self.Ws, self.bs)):
             self.Ms[i] = M.to(*args, **kwargs)

madminer/utils/ml/models/made_mog.py

@@ -201,7 +201,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(ConditionalMixtureMADE, self).to(*args, **kwargs)
 
         for i, M in enumerate(self.Ms):
             self.Ms[i] = M.to(*args, **kwargs)

madminer/utils/ml/models/maf.py

@@ -103,7 +103,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(MaskedAutoregressiveFlow, self).to(*args, **kwargs)
 
         for i, (made) in enumerate(self.mades):
             self.mades[i] = made.to(*args, **kwargs)
@@ -260,7 +260,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(ConditionalMaskedAutoregressiveFlow, self).to(*args, **kwargs)
 
         for i, (made) in enumerate(self.mades):
             self.mades[i] = made.to(*args, **kwargs)

madminer/utils/ml/models/maf_mog.py

@@ -137,7 +137,7 @@ def to(self, *args, **kwargs):
         self.to_args = args
         self.to_kwargs = kwargs
 
-        self = super().to(*args, **kwargs)
+        self = super(ConditionalMixtureMaskedAutoregressiveFlow, self).to(*args, **kwargs)
 
         for i, (made) in enumerate(self.mades):
             self.mades[i] = made.to(*args, **kwargs)