Shrink step length for each resampling iteration

popt/update_schemes/enopt.py

@@ -172,11 +172,13 @@ def calc_update(self):
         improvement = False
         success = False
         resampling_iter = 0
+        self.optimizer.restore_parameters()
 
         while not improvement:  # resampling loop
 
-            # Shrink covariance each time we try resampling
+            # Shrink covariance and step size each time we try resampling
             shrink = self.cov_factor ** resampling_iter
+            self.optimizer.apply_backtracking(np.sqrt(self.cov_factor)** resampling_iter) 
 
             # Calculate gradient
             if self.nesterov:

popt/update_schemes/genopt.py

@@ -138,6 +138,7 @@ def calc_update(self):
         improvement = False
         success = False
         resampling_iter = 0
+        self.optimizer.restore_parameters()
 
         while improvement is False:  # resampling loop
 

popt/update_schemes/smcopt.py

@@ -126,11 +126,13 @@ def calc_update(self,):
         success = False
         resampling_iter = 0
         inflate = 2 * (self.inflation_factor + self.iteration)
+        self.optimizer.restore_parameters()
 
         while improvement is False:  # resampling loop
 
-            # Shrink covariance each time we try resampling
+            # Shrink covariance and step size each time we try resampling
             shrink = self.cov_factor ** resampling_iter
+            self.optimizer.apply_backtracking(np.sqrt(self.cov_factor) ** resampling_iter)
 
             # Calc sensitivity
             (sens_matrix, self.best_state, best_func_tmp) = self.sens(self.mean_state, inflate,

popt/update_schemes/subroutines/optimizers.py

@@ -124,12 +124,12 @@ def apply_smc_update(self, control, gradient, **kwargs):
         new_control = (1-alpha) * control + alpha * gradient
         return new_control
 
-    def apply_backtracking(self):
+    def apply_backtracking(self, shrink=0.5):
         """
         Apply backtracking by reducing step size and momentum temporarily.
         """
-        self._step_size = 0.5*self._step_size
-        self._momentum  = 0.5*self._momentum
+        self._step_size = shrink*self._step_size
+        self._momentum  = shrink*self._momentum
 
     def restore_parameters(self):
         """