Changes random seed behaviour and includes likelihood logging

On every call of `pipeline._core_likelihood` the seeds are reset using the `pipeline._randomness()` method. This allows the same seeds to be used for the stochastic fields on every point in the parameter space (except for the free case).
Basic logging, which prints the value of the liklihood on every evaluation, can be achieved using `logging.basicConfig(filename=FOO/BAR,level=logging.INFO)`.

imagine/pipelines/pipeline.py

@@ -149,6 +149,8 @@ def chains_directory(self, chains_directory):
 
     def clean_chains_directory(self):
         """Removes the contents of the chains directory"""
+        log.debug('@ pipeline::clean_chains_directory')
+
         if mpirank==0:
             for f in os.listdir(self.chains_directory):
                 fullpath = path.join(self.chains_directory, f)
@@ -477,6 +479,8 @@ def tidy_up(self):
         """
         Resets internal state before a new run
         """
+        log.debug('@ pipeline::tidy_up')
+
         self.results = None
         self._evidence = None
         self._evidence_err = None
@@ -506,6 +510,7 @@ def _randomness(self):
         else:
             self.ensemble_seeds = None
 
+
     # This function returns all parameter names of all factories in order
     def get_par_names(self):
         # Create list of names
@@ -561,6 +566,9 @@ def _core_likelihood(self, cube):
         log.debug('@ pipeline::_core_likelihood')
         log.debug('sampler at %s' % str(cube))
 
+        # Resets the seeds
+        self._randomness()
+
         # Obtain observables for provided cube
         observables = self._get_observables(cube)
 
@@ -569,6 +577,9 @@ def _core_likelihood(self, cube):
         # check likelihood value until negative (or no larger than given threshold)
         if self.check_threshold and current_likelihood > self.likelihood_threshold:
             raise ValueError('log-likelihood beyond threshold')
+
+        log.info('Likelihood evaluation at point:'
+                 ' {0} value: {1}'.format(cube, current_likelihood))
         return current_likelihood * self.likelihood_rescaler
 
     def _mpi_likelihood(self, cube):