Add example of IPOP-CMA-ES

.github/workflows/examples.yml

@@ -26,4 +26,5 @@ jobs:
           pip install --progress-bar off optuna
           pip install --progress-bar off -U .
       - run: python examples/quadratic_function.py
+      - run: python examples/ipop_cmaes.py
       - run: python examples/optuna_sampler.py

examples/ipop_cmaes.py

@@ -0,0 +1,38 @@
+import numpy as np
+from cmaes import CMA
+
+
+def quadratic(x1, x2):
+    return (x1 - 3) ** 2 + (10 * (x2 + 2)) ** 2
+
+
+def main():
+    cma_opts = {"mean": np.zeros(2), "sigma": 1.3}
+    optimizer = CMA(**cma_opts)
+
+    # Multiplier for increasing population size before each restart.
+    inc_popsize = 2
+
+    print(" g    f(x1,x2)     x1      x2  ")
+    print("===  ==========  ======  ======")
+    for generation in range(150):
+        solutions = []
+        for _ in range(optimizer.population_size):
+            x = optimizer.ask()
+            value = quadratic(x[0], x[1])
+            solutions.append((x, value))
+
+            msg = "{g:3d}  {value:10.5f}  {x1:6.2f}  {x2:6.2f}".format(
+                g=generation, value=value, x1=x[0], x2=x[1],
+            )
+            print(msg)
+        optimizer.tell(solutions)
+
+        if optimizer.should_terminate():
+            popsize = optimizer.population_size * inc_popsize
+            optimizer = CMA(population_size=popsize, **cma_opts)
+            print("Restart CMA-ES with popsize={}".format(popsize))
+
+
+if __name__ == "__main__":
+    main()

examples/quadratic_function.py

@@ -11,19 +11,22 @@ def main():
     print(" g    f(x1,x2)     x1      x2  ")
     print("===  ==========  ======  ======")
 
-    for generation in range(50):
+    while True:
         solutions = []
         for _ in range(optimizer.population_size):
             x = optimizer.ask()
             value = quadratic(x[0], x[1])
             solutions.append((x, value))
 
             msg = "{g:3d}  {value:10.5f}  {x1:6.2f}  {x2:6.2f}".format(
-                g=generation, value=value, x1=x[0], x2=x[1],
+                g=optimizer.generation, value=value, x1=x[0], x2=x[1],
             )
             print(msg)
         optimizer.tell(solutions)
 
+        if optimizer.should_terminate():
+            break
+
 
 if __name__ == "__main__":
     main()