Merge branch 'main' of https://github.com/MDCHAMP/opt2 into main

src/freelunch/base.py

@@ -23,7 +23,7 @@ def __init__(self, obj=None, hypers={}, bounds=None):
         Unlikely to instance the base class but idk what else goes here
         '''
         self.hypers = dict(self.hyper_defaults, **hypers) # Hyperparamters/ methods (dictionary of )
-        self.bounds = bounds # Bounds / constraints
+        self.bounds = np.array(bounds) # Bounds / constraints
         self.nfe = 0
         self.obj = self.wrap_obj_with_nfe(obj) # Objective function 
 
@@ -86,8 +86,11 @@ def wrap_obj_with_nfe(self, obj):
         if obj is None: return None
         def w_obj(vec):
             self.nfe +=1
-            # TODO: inf, nan and None handling 
-            return obj(vec)
+            fit = obj(vec)
+            try:
+                return float(fit)
+            except(ValueError, TypeError):
+                return None
         return w_obj
 
 # Subclasses for granularity

src/freelunch/darwin.py

@@ -135,14 +135,9 @@ class binary_tournament(genetic_operation):
     def op(self, olds, news):
         out = np.empty_like(olds, dtype=object)
         for old, new, i in zip(olds, news, range(len(out))):
-            if new.fitness < old.fitness:
+            if new < old:
                 out[i] = new
                 new.on_win()
-            elif old.fitness <= new.fitness:
-                out[i] = old
             else:
-                raise BadObjectiveFunctionScores(
-                    'Winner could not be determined by comparing objective scores. scores:{} and {}'.format(
-                        old.fitness, new.fitness
-                    ))
+                out[i] = old
         return out

src/freelunch/optimisers.py

@@ -185,7 +185,7 @@ def run(self):
             for i, o, n in zip(range(self.hypers['N']), old, new):
                 if self.P(o.fitness, n.fitness, T) >= np.random.uniform(0,1): 
                     old[i] = new[i]
-                if n.fitness < best.fitness:
+                if n < best:
                     best = n
         return np.array([best])
 

src/freelunch/tech.py

@@ -37,8 +37,6 @@ def uniform_continuous_init(bounds, N, creature=animal):
 def compute_obj(pop, obj):
     for sol in pop:
         sol.fitness = obj(sol.dna)
-        if np.isnan(sol.fitness):
-            sol.fitness = None
     return pop
 
 

src/freelunch/zoo.py

@@ -37,7 +37,18 @@ def on_win(self):
                 t.win()
         self.tech = []
 
-
+    def __lt__(self, other):
+        '''overlaod the < operator for convenient handling of tournament selction in the presence onf nonetype fitnesses'''
+        if self.fitness is None:
+            if other.fitness is None:
+                # ? what to do here
+                return False
+            return False # Other has lower fitness
+        elif other.fitness is None:
+            return True # We have the lower fitness            
+        else:
+            return self.fitness < other.fitness
+
 # %% All that the light touches is our domain
 
 

tests/test_opt.py

@@ -27,7 +27,6 @@ def test_base_optimiser():
 	opt.obj = lambda x:x
 	opt(nruns=2)
 
-@pytest.mark.skip(reason="Objective sanitation needs propper handling")
 def test_naughty_objective():
 	out = DE(obj=naughty_objective, bounds=[[-1, 1]])(nruns=1, full_output=True)