feat: add new parameters for DECC

- Add new boolean flag to ensure the number of evaluations used
  (default=False);
- Update version in pyproject.toml;
- Add number of evaluations in DECC output;

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "decc"
-version = "0.2.2"
+version = "0.2.3"
 readme = "README.md"
 description = "Cooperative Co-evolutionary Differential Evolution algorithms in Python."
 

src/decc/optimizers/decc.py

@@ -27,13 +27,15 @@ def __init__(self,
                  grouping: Literal['halve',
                                    'dims'] = 'halve',
                  F: float = 0.8,
-                 CR: float = 0.3) -> None:
+                 CR: float = 0.3,
+                 ensure_evaluations: bool = False) -> None:
         super().__init__(problem, seed)
         self.subpop_size = subpopulation_size
         self.F = F
         self.CR = CR
         self.max_fn = max_fn
         self.variant = 'DECC-'
+        self.ensure_evaluations = ensure_evaluations
 
         d = self.problem.dims
         if grouping == 'halve':
@@ -55,7 +57,7 @@ def parameters(self) -> dict:
             'max_evaluations': self.max_fn,
             'n_subproblems': len(self.subproblem_indices)
         }
-    
+
     def name(self) -> str:
         return self.variant
 
@@ -68,7 +70,6 @@ def _optimize(self, *args, **kwargs) -> tuple[np.ndarray,
         rng = np.random.default_rng(self.seed)
         l, u = self.problem.bounds
         d = self.problem.dims
-        p = self.pop_size
         sp = self.subpop_size
         fn = self.problem.fn
         n_evaluations = 0
@@ -125,8 +126,13 @@ def update_best(population: np.ndarray,
             subpopulations.append(population)
             subpopulations_fitness.append(population_fitness)
 
+            # Early stop
+            if self.ensure_evaluations and \
+                    n_evaluations >= self.max_fn:
+                break
+
         # Evolution loop
-        while n_evaluations <= self.max_fn:
+        while n_evaluations < self.max_fn:
             # Obtaining the new context vector
             context_vector = np.zeros((d,),
                                       dtype=np.float32)
@@ -185,7 +191,14 @@ def _fn(pop: np.ndarray) -> np.ndarray:
                 update_best(subpopulations[i],
                             subpopulations_fitness[i])
 
-        return best_fitness.squeeze(), best_solution, None
+                # Early stop
+                if self.ensure_evaluations and \
+                        n_evaluations >= self.max_fn:
+                    break
+
+        return (best_fitness.squeeze(),
+                best_solution,
+                dict(n_evaluations=n_evaluations))
 
     def _population_w_context(self,
                               context: np.ndarray,

src/decc/optimizers/decc_g.py

@@ -30,7 +30,8 @@ def __init__(self,
                  max_fn: int = int(1e6),
                  F: float = 0.8,
                  CR: float = 0.3,
-                 weights_bound: tuple[float, float] = (-5, 5)) -> None:
+                 weights_bound: tuple[float, float] = (-5, 5),
+                 ensure_evaluations: bool = False) -> None:
         super().__init__(problem, seed)
         assert n_subproblems <= self.problem.dims
 
@@ -42,6 +43,7 @@ def __init__(self,
         self.de_eval = de_evaluations
         self.sansde_eval = sansde_evaluations
         self.w_bounds = weights_bound
+        self.ensure_evaluations = ensure_evaluations
 
     def parameters(self) -> dict:
         return {
@@ -57,7 +59,7 @@ def parameters(self) -> dict:
             'max_evaluations': self.max_fn,
             'n_subproblems': self.n_sub
         }
-    
+
     def name(self) -> str:
         return "DECC-G"
 
@@ -73,8 +75,8 @@ def _optimize(self, *args, **kwargs) -> tuple[np.ndarray,
         p = self.pop_size
         fn = self.problem.fn
         n_evaluations = 0
-        best_solution = None
-        best_fitness = None
+        best_solution: np.ndarray = None
+        best_fitness: np.ndarray = None
         seed_interval = (0, 99999)
 
         # Random initial population of
@@ -104,7 +106,7 @@ def update_best(population: np.ndarray,
         # Update best
         update_best(population, population_fitness)
 
-        while n_evaluations <= self.max_fn:
+        while n_evaluations < self.max_fn:
             # Generating random groups
             group_seed = rng.integers(*seed_interval)
             groups = decomposition.random_group_decompose(
@@ -149,6 +151,11 @@ def _fn(pop: np.ndarray) -> np.ndarray:
                 # Update best
                 update_best(population, population_fitness)
 
+                # Early stop
+                if self.ensure_evaluations and \
+                        n_evaluations >= self.max_fn:
+                    break
+
         # === Updating weight population ===
         def _fn(w_pop: np.ndarray,
                 idx: int):
@@ -184,6 +191,11 @@ def _update_pop_w_weights(idx: int):
             nonlocal n_evaluations
             nonlocal population
 
+            # Early stop
+            if self.ensure_evaluations and \
+                    n_evaluations >= self.max_fn:
+                return
+
             # Obtain initial fitness of
             #   weighted individuals
             w_f = _fn(weight_population, idx)
@@ -229,4 +241,6 @@ def _update_pop_w_weights(idx: int):
         # Update best
         update_best(population, population_fitness)
 
-        return best_fitness.squeeze(), best_solution, None
+        return (best_fitness.squeeze(),
+                best_solution,
+                dict(n_evaluations=n_evaluations))