CMA variants by Risto

nevergrad/optimization/experimentalvariants.py

@@ -354,3 +354,43 @@
 # High-Speed variants
 HSDE = DifferentialEvolution(high_speed=True).set_name("HSDE", register=True)
 LhsHSDE = DifferentialEvolution(initialization="LHS", high_speed=True).set_name("LhsHSDE", register=True)
+
+# CMA variants built by Risto.
+# The four configurations obtained on yasmallbbob (CMA1 is supposed to be the best).
+# 49  --solver=CMA-ES --scale=0.4151 --popsize_factor=9 --elitist=False --diagonal=False --fcmaes=True --random_init=False
+# 35  --solver=CMA-ES --scale=0.3177 --popsize_factor=7 --elitist=False --diagonal=False --fcmaes=True --random_init=False
+# 10  --solver=CMA-ES --scale=6.1404 --popsize_factor=8 --elitist=True --diagonal=False --fcmaes=False --random_init=False
+# 60  --solver=CMA-ES --scale=0.4267 --popsize_factor=9 --elitist=False --diagonal=False --fcmaes=True --random_init=False
+CMA1 = ParametrizedCMA(scale=0.4151, popsize_factor=9, elitist=False, diagonal=False, fcmaes=True).set_name(
+    "CMA1", register=True
+)
+CMA2 = ParametrizedCMA(scale=0.3177, popsize_factor=7, elitist=False, diagonal=False, fcmaes=True).set_name(
+    "CMA2", register=True
+)
+CMA3 = ParametrizedCMA(scale=6.1404, popsize_factor=8, elitist=True, diagonal=False, fcmaes=False).set_name(
+    "CMA3", register=True
+)
+CMA4 = ParametrizedCMA(scale=0.4267, popsize_factor=9, elitist=False, diagonal=False, fcmaes=True).set_name(
+    "CMA4", register=True
+)
+# The four configurations obtained on yabbob.
+# 51  --solver=CMA-ES --scale=0.3739 --popsize_factor=3 --elitist=False --diagonal=False --fcmaes=False
+# --random_init=False
+# 45  --solver=CMA-ES --scale=0.3607 --popsize_factor=3 --elitist=False --diagonal=False --fcmaes=False
+##--random_init=False
+# 44  --solver=CMA-ES --scale=0.4699 --popsize_factor=3 --elitist=False --diagonal=False --fcmaes=False
+# --random_init=False
+# 27  --solver=CMA-ES --scale=0.6221 --popsize_factor=4 --elitist=False --diagonal=False --fcmaes=False
+# --random_init=False
+CMA5 = ParametrizedCMA(scale=0.3739, popsize_factor=3, elitist=False, diagonal=False, fcmaes=False).set_name(
+    "CMA5", register=True
+)
+CMA6 = ParametrizedCMA(scale=0.3607, popsize_factor=3, elitist=False, diagonal=False, fcmaes=False).set_name(
+    "CMA6", register=True
+)
+CMA7 = ParametrizedCMA(scale=0.4699, popsize_factor=3, elitist=False, diagonal=False, fcmaes=False).set_name(
+    "CMA7", register=True
+)
+CMA8 = ParametrizedCMA(scale=0.6221, popsize_factor=4, elitist=False, diagonal=False, fcmaes=False).set_name(
+    "CMA8", register=True
+)

nevergrad/optimization/optimizerlib.py

@@ -419,8 +419,11 @@ def __init__(
         super().__init__(parametrization, budget=budget, num_workers=num_workers)
         self._config = ParametrizedCMA() if config is None else config
         pop = self._config.popsize
-        self._popsize = max(num_workers, 4 + int(3 * np.log(self.dimension))) if pop is None else pop
-        # internal attributes
+        self._popsize = (
+            max(num_workers, 4 + int(self._config.popsize_factor * np.log(self.dimension)))
+            if pop is None
+            else pop
+        )  # internal attributes
         self._to_be_asked: tp.Deque[np.ndarray] = deque()
         self._to_be_told: tp.List[p.Parameter] = []
         self._num_spawners = self._popsize // 2  # experimental, for visualization
@@ -532,6 +535,8 @@ class ParametrizedCMA(base.ConfiguredOptimizer):
     popsize: Optional[int] = None
         population size, should be n * self.num_workers for int n >= 1.
         default is max(self.num_workers, 4 + int(3 * np.log(self.dimension)))
+    popsize_factor: float = 3.
+        factor in the formula for computing the population size
     diagonal: bool
         use the diagonal version of CMA (advised in big dimension)
     high_speed: bool
@@ -554,6 +559,7 @@ def __init__(
         scale: float = 1.0,
         elitist: bool = False,
         popsize: tp.Optional[int] = None,
+        popsize_factor: float = 3.0,
         diagonal: bool = False,
         high_speed: bool = False,
         fcmaes: bool = False,
@@ -567,6 +573,7 @@ def __init__(
         self.scale = scale
         self.elitist = elitist
         self.popsize = popsize
+        self.popsize_factor = popsize_factor
         self.diagonal = diagonal
         self.fcmaes = fcmaes
         self.high_speed = high_speed

nevergrad/optimization/recorded_recommendations.csv

@@ -13,6 +13,14 @@ AvgRandomSearch,1.7163318206,-1.0929434198,-0.5715059829,0.2691248199,,,,,,,,,,,
 BPRotationInvariantDE,-0.5273201105,-0.156805946,-1.2551455501,3.327306828,2.3993996197,1.2830831401,0.8566151549,-1.6450076557,,,,,,,,
 CM,1.0082049151,-0.9099785499,-1.025147209,1.2046460074,,,,,,,,,,,,
 CMA,1.012515477,-0.9138805701,-1.029555946,1.2098418178,,,,,,,,,,,,
+CMA1,0.4202951745,-0.3793470828,-0.4273579891,0.5021865062,,,,,,,,,,,,
+CMA2,0.321676167,-0.2903362279,-0.3270817469,0.384352332,,,,,,,,,,,,
+CMA3,3.0816966941,0.8525817234,3.9346279048,3.2381549254,,,,,,,,,,,,
+CMA4,0.432040354,-0.3899479649,-0.4393005395,0.5162201451,,,,,,,,,,,,
+CMA5,0.3785795368,-0.3416999452,-0.3849509682,0.4523598557,,,,,,,,,,,,
+CMA6,0.3652143325,-0.3296367217,-0.3713608297,0.4363899437,,,,,,,,,,,,
+CMA7,0.4757810226,-0.4294324799,-0.483788339,0.5685046702,,,,,,,,,,,,
+CMA8,0.6298858782,-0.5685251027,-0.640486754,0.7526425949,,,,,,,,,,,,
 CMandAS2,1.012515477,-0.9138805701,-1.029555946,1.2098418178,,,,,,,,,,,,
 CMandAS3,0.4729858315,-0.6814343973,0.2424455578,-1.7007994128,,,,,,,,,,,,
 CauchyLHSSearch,-0.527971877,1.341890246,2.6790716005,3.5963545262,,,,,,,,,,,,