update

tensiometer/synthetic_probability/synthetic_probability.py

@@ -494,8 +494,12 @@ def _init_trainable_bijector(self, trainable_bijector, trainable_bijector_path=N
         # load from file:
         if trainable_bijector_path is not None:
             if self.trainable_transformation is not None:
+                if self.feedback > 1:
+                    print('    - loading trainable bijector from file:', trainable_bijector_path)
                 self.trainable_transformation = self.trainable_transformation.load(trainable_bijector_path, 
                                                                                    **kwargs)
+            else:
+                raise ValueError('Cannot load a bijector from file if the trainable bijector is not a TrainableTransformation')
 
         # initialize bijector:
         if self.trainable_transformation is not None:
@@ -1029,11 +1033,23 @@ def MCSamples(self, size, logLikes=True, **kwargs):
         :param size: number of samples
         :param logLikes: logical, whether to include log-likelihoods or not.
         """
+        # sample:
         samples = self.sample(size)
+        finite_filter = tf.reduce_all(tf.math.is_finite(samples), axis=-1)
         if logLikes:
             loglikes = -self.log_probability(samples)
+            finite_filter = tf.math.logical_and(finite_filter, tf.math.is_finite(loglikes))
         else:
             loglikes = None
+        # filter out non-finite values:
+        if not np.all(finite_filter):
+            samples = samples[finite_filter]
+            if loglikes is not None:
+                loglikes = loglikes[finite_filter]        
+            # feedback:
+            if self.feedback > 0:
+                print('    - found non-finite values, filtering out {0} samples'.format(size - len(samples)))
+        # create MCSamples object:
         mc_samples = MCSamples(
             samples=samples.numpy(),
             loglikes=loglikes.numpy(),
@@ -2353,6 +2369,8 @@ def training_plot(self, logs=None, file_path=None, ipython_plotting=False):
                                file_path=_temp_name, 
                                ipython_plotting=ipython_plotting,
                                title='Training flow '+str(_i))
+            if ipython_plotting:
+                plt.show()
         #
         return None
 
@@ -2574,7 +2592,7 @@ def average_flow_from_chain(chain, num_flows=1, cache_dir=None, root_name='sprob
             if use_mpi and size > 1:
                 pass
             else:
-                if feedback > 1:
+                if feedback > 0:
                     print('Loading flow', i, 'from cache', flush=use_mpi)
                 flow = FlowCallback.load(chain, _outroot, **kwargs)
                 flows.append(flow)