Merge pull request #2 from Leviathan321/custom-ops

Make possible to use custom operators.

opensbt/algorithm/nsga2_optimizer.py

@@ -1,3 +1,4 @@
+from opensbt.utils.operators import select_operator
 from pymoo.core.problem import Problem
 from pymoo.termination import get_termination
 from pymoo.algorithms.moo.nsga2 import NSGA2
@@ -17,7 +18,6 @@ class NsgaIIOptimizer(Optimizer):
 
     def __init__(self,
                 problem: Problem,
-
                 config: SearchConfiguration):
         self.config = config
         self.problem = problem
@@ -40,11 +40,12 @@ def __init__(self,
         self.algorithm = NSGA2(
             pop_size=config.population_size,
             n_offsprings=config.num_offsprings,
-            sampling=FloatRandomSampling(),
-            crossover=SBX(prob=config.prob_crossover, eta=config.eta_crossover),
-            mutation=PM(prob=config.prob_mutation, eta=config.eta_mutation),
-            eliminate_duplicates=True)
-
+            sampling = select_operator("init", config),
+            crossover = select_operator("cx", config),
+            mutation = select_operator("mut", config),
+            eliminate_duplicates = select_operator("dup", config)
+        )
+
         ''' Prioritize max search time over set maximal number of generations'''
         if config.maximal_execution_time is not None:
             self.termination = get_termination("time", config.maximal_execution_time)

opensbt/algorithm/nsga2dt_optimizer.py

@@ -1,6 +1,7 @@
 import random
 from opensbt.model_ga.result  import SimulationResult
 from opensbt.evaluation.critical import Critical
+from opensbt.utils.operators import select_operator
 from pymoo.termination import get_termination
 from pymoo.algorithms.moo.nsga2 import NSGA2
 from pymoo.core.population import Population
@@ -125,10 +126,11 @@ def run(self) -> SimulationResult:
         inner_algorithm = NSGA2(
             pop_size=None,
             n_offsprings=None,
-            sampling=None,
-            crossover=SBX(prob=prob_crossover, eta=eta_crossover),
-            mutation=PM(prob=prob_mutation, eta=eta_mutation),
-            eliminate_duplicates=True)
+            sampling = select_operator("init", config),
+            crossover = select_operator( "cx", config),
+            mutation = select_operator( "mut", config),
+            eliminate_duplicates = select_operator( "dup", config)
+        )
 
         tree_iteration = 0
         n_func_evals = 0
@@ -153,9 +155,10 @@ def run(self) -> SimulationResult:
                     pop_size=pop_size,
                     n_offsprings=num_offsprings,
                     sampling=initial_population,
-                    crossover=SBX(prob=prob_crossover, eta=eta_crossover),
-                    mutation=PM(prob=prob_mutation, eta=eta_mutation),
-                    eliminate_duplicates=True)
+                    crossover = select_operator( "cx", config),
+                    mutation = select_operator( "mut", config),
+                    eliminate_duplicates = select_operator( "dup", config)
+                )
 
                 termination = get_termination("n_gen", inner_num_gen)
 

opensbt/algorithm/pso_optimizer.py

@@ -1,3 +1,4 @@
+from opensbt.utils.operators import select_operator
 from pymoo.core.problem import Problem
 from pymoo.termination import get_termination
 from pymoo.algorithms.soo.nonconvex.pso import PSO
@@ -38,7 +39,8 @@ def __init__(self,
 
         # initialize algorithm
         self.algorithm = PSO(
-            pop_size=config.population_size,
+            pop_size = config.population_size,
+            sampling = select_operator("init",config)
         )
 
         ''' Prioritize max search time over set maximal number of generations'''

opensbt/experiment/search_configuration.py

@@ -1,4 +1,7 @@
+from dataclasses import dataclass
 
+
+@dataclass
 class SearchConfiguration(object):
     """ This class holds all configuration parameter related to opimization algorithms
     """
@@ -24,6 +27,16 @@ class SearchConfiguration(object):
 
     seed = None
 
+    operators= {
+        "cx" : None,
+        "mut" : None,
+        "dup" : None,
+        "init" : None
+    }
+
+    custom_params = {  # to be forwarded to operators
+
+    }
 
 class DefaultSearchConfiguration(SearchConfiguration):
     """ This class holds all configuration parameter initialized with default values 
@@ -48,4 +61,15 @@ class DefaultSearchConfiguration(SearchConfiguration):
     ref_point_hv = None
     nadir = ref_point_hv
 
-    seed = None
+    seed = None
+
+    operators= {
+        "cx" : None,
+        "mut" : None,
+        "dup" : None,
+        "init" : None
+    }
+
+    custom_params = {  # to be forwarded to operators
+
+    }

opensbt/utils/operators.py

@@ -0,0 +1,33 @@
+from pymoo.operators.crossover.sbx import SBX # type: ignore
+from pymoo.operators.mutation.pm import PM # type: ignore
+from pymoo.operators.sampling.lhs import LHS # type: ignore
+
+def select_operator(operation,
+                    config, 
+                    **kwargs):
+    """
+    Selects either the default operator or a custom operator based on the condition.
+    """
+    if kwargs is not None:
+        kwargs = {}
+
+    if config.operators[operation] is None:
+        if operation == "mut":
+            operator = PM   
+            if "prob" not in kwargs:
+                kwargs["prob"] = config.prob_mutation
+            if "eta" not in kwargs:
+                kwargs["eta"] = config.eta_mutation
+        elif operation == "cx":
+            operator = SBX
+            if "prob" not in kwargs:
+                kwargs["prob"] = config.prob_crossover
+            if "eta" not in kwargs:
+                kwargs["eta"] = config.eta_crossover
+        elif operation == "init":
+            operator = LHS
+        elif operation == "dup":
+            return True
+    else:
+        operator = config.operators[operation]
+    return operator(**kwargs)  # Passes the keyword arguments to the operator