[Opt] Wrapped DifferentialEvolution and DualAnnealing from scipy

src/jax_fdm/optimization/optimizers/__init__.py

@@ -2,6 +2,8 @@
 from .constrained import *  # noqa F403
 from .second_order import *  # noqa F403
 from .gradient_based import *  # noqa F403
+from .gradient_free import *  # noqa F403
+from .evolutionary import *  # noqa F403
 from .ipopt import *  # noqa F403
 
 

src/jax_fdm/optimization/optimizers/evolutionary.py

@@ -0,0 +1,96 @@
+"""
+A collection of evolutionary optimizers.
+"""
+from jax import vmap
+
+from scipy.optimize import differential_evolution
+from scipy.optimize import dual_annealing
+
+from jax_fdm.optimization.optimizers import Optimizer
+
+
+# ==========================================================================
+# Optimizers
+# ==========================================================================
+
+class DifferentialEvolution(Optimizer):
+    """
+    The a differential evolution optimizer with box constraints.
+    """
+    def __init__(self, popsize=20, vectorized=False, num_workers=1, seed=43, display=False, **kwargs):
+        super().__init__(name="DifferentialEvolution", disp=display, **kwargs)
+        self.popsize = popsize
+        self.vectorized = vectorized
+        self.num_workers = num_workers
+        self.seed = seed
+
+    def _minimize(self, opt_problem):
+        """
+        Scipy backend method to minimize a loss function.
+        """
+        fun = opt_problem["fun"]
+
+        def func(x):
+            return fun(x)[0]
+
+        def func_vmap(x):
+            result = vmap(func, in_axes=(1))(x)
+            return result
+
+        _args = None
+        opt_problem["func"] = func
+        if self.vectorized:
+            opt_problem["func"] = func_vmap
+
+        opt_problem["vectorized"] = self.vectorized
+        opt_problem["polish"] = False
+        opt_problem["seed"] = self.seed
+
+        opt_problem["popsize"] = self.popsize
+        opt_problem["maxiter"] = opt_problem["options"]["maxiter"]
+        opt_problem["disp"] = opt_problem["options"]["disp"]
+        opt_problem["args"] = _args
+        opt_problem["updating"] = "deferred" if self.vectorized else "immediate"
+        opt_problem["workers"] = self.num_workers
+
+        del opt_problem["fun"]
+        del opt_problem["jac"]
+        del opt_problem["hess"]
+        del opt_problem["method"]
+        del opt_problem["options"]
+
+        return differential_evolution(**opt_problem)
+
+
+class DualAnnealing(Optimizer):
+    """
+    The a dual annealing optimizer with box constraints.
+    """
+    def __init__(self, no_local_search=True, seed=None, display=False, **kwargs):
+        super().__init__(name="DualAnnealing", disp=display, **kwargs)
+        self.no_local_search = no_local_search
+        self.seed = seed
+
+    def _minimize(self, opt_problem):
+        """
+        Scipy backend method to minimize a loss function.
+        """
+        fun = opt_problem["fun"]
+
+        def func(x, *args, **kwargs):
+            return fun(x)[0]
+
+        opt_problem["func"] = func
+        opt_problem["no_local_search"] = self.no_local_search
+        opt_problem["maxiter"] = opt_problem["options"]["maxiter"]
+        opt_problem["args"] = (None, )
+
+        del opt_problem["tol"]
+        del opt_problem["fun"]
+        del opt_problem["jac"]
+        del opt_problem["hess"]
+        del opt_problem["constraints"]
+        del opt_problem["method"]
+        del opt_problem["options"]
+
+        return dual_annealing(**opt_problem)