Improved bounds check to deal with non-numericals, other improvements

Author: fjwillemsen

kernel_tuner/strategies/common.py

@@ -242,7 +242,11 @@ def get_bounds(self):
         """Create a bounds array from the tunable parameters."""
         bounds = []
         for values in self.encoded_params_values if self.encode_non_numeric else self.searchspace.params_values:
-            bounds.append((min(values), max(values)))
+            try:
+                bounds.append((min(values), max(values)))
+            except TypeError:
+                # if values are not numeric, use the first and last value as bounds
+                bounds.append((values[0], values[-1]))
         return bounds
 
     def encoded_to_params(self, config):

kernel_tuner/strategies/diff_evo.py

@@ -45,7 +45,7 @@ def tune(searchspace: Searchspace, runner, tuning_options):
 
     try:
         differential_evolution(searchspace, cost_func, bounds, popsize, maxiter, F, CR, method, constraint_aware, tuning_options.verbose)
-    except util.StopCriterionReached as e:
+    except StopCriterionReached as e:
         if tuning_options.verbose:
             print(e)
 

kernel_tuner/strategies/pso.py

@@ -23,18 +23,11 @@ def tune(searchspace: Searchspace, runner, tuning_options):
     # scale variables in x because PSO works with velocities to visit different configurations
     cost_func = CostFunc(searchspace, tuning_options, runner, scaling=True)
 
-<<<<<<< HEAD
     # using this instead of get_bounds because scaling is used
-    bounds, _, eps = cost_func.get_bounds_x0_eps()
+    bounds, x0, eps = cost_func.get_bounds_x0_eps()
 
     num_particles, maxiter, w, c1, c2, constraint_aware = common.get_options(tuning_options.strategy_options, _options)
     num_particles = min(round(searchspace.size / 2), num_particles)
-=======
-    #using this instead of get_bounds because scaling is used
-    bounds, x0, eps = cost_func.get_bounds_x0_eps()
-
-    num_particles, maxiter, w, c1, c2 = common.get_options(tuning_options.strategy_options, _options)
->>>>>>> origin/custom_diff_evo
 
     best_score_global = sys.float_info.max
     best_position_global = []

kernel_tuner/strategies/simulated_annealing.py

@@ -35,13 +35,8 @@ def tune(searchspace: Searchspace, runner, tuning_options):
     max_fevals = min(searchspace.size, max_fevals)
 
     # get random starting point and evaluate cost
-<<<<<<< HEAD
-    pos = generate_starting_point(searchspace, constraint_aware)
-    old_cost = cost_func(pos, check_restrictions=not constraint_aware)
-=======
     pos = cost_func.get_start_pos()
-    old_cost = cost_func(pos, check_restrictions=False)
->>>>>>> origin/custom_diff_evo
+    old_cost = cost_func(pos, check_restrictions=not constraint_aware)
 
     # main optimization loop
     stuck = 0