adding_plotting_curve (#1564)

* adding_plotting_curve

* addgomea (#1565)

* addgomea

* fix_missing_import

* fix_missing_import

* fix_missing_import

* fix_missing_import

* fix_missing_import

* fix_missing_import

* Update test_optimizerlib.py

* fix

* fix

* fix

Author: teytaud

examples/example_for_plotting_optim_curve.py

@@ -0,0 +1,7 @@
+import nevergrad as ng
+
+def f(x):
+   return sum((x-1.234)**2)
+opt = ng.optimizers.TwoPointsDE(3, 500)
+opt.minimize(f)
+print(opt.optim_curve)

mypy.ini

@@ -1,7 +1,7 @@
 [mypy]
 
 #[mypy-scipy.*,requests,smac,smac.*,pandas,compiler_gym,compiler_gym.*,gym,gym.*,gym_anm,matplotlib.*,pytest,cma,bayes_opt.*,torchvision.models,torch.*,mpl_toolkits.*,fcmaes.*,tqdm,pillow,PIL,PIL.Image,sklearn.*,pyomo.*,pyproj,IOHexperimenter.*,tensorflow,koncept.models,cv2,imquality,imquality.brisque,lpips,mixsimulator.*,networkx.*,cdt.*,]
-[mypy-scipy.*,requests,pandas,compiler_gym,compiler_gym.*,gym_anm,matplotlib.*,pytest,cma,bayes_opt.*,torchvision.models,torch.*,mpl_toolkits.*,fcmaes.*,tqdm,pillow,PIL,PIL.Image,sklearn.*,pyomo.*,pyproj,IOHexperimenter.*,tensorflow,koncept.models,cv2,imquality,imquality.brisque,lpips,mixsimulator.*,networkx.*,cdt.*,pymoo,pymoo.*,bayes_optim.*,olympus.*,pymoo,pymoo.*,pybullet,pybullet_envs,pybulletgym,pyvirtualdisplay,nlopt,aquacrop.*]
+[mypy-scipy.*,requests,pandas,compiler_gym,compiler_gym.*,gym_anm,matplotlib.*,pytest,cma,bayes_opt.*,torchvision.models,torch.*,mpl_toolkits.*,fcmaes.*,tqdm,pillow,PIL,PIL.Image,sklearn.*,pyomo.*,pyproj,IOHexperimenter.*,tensorflow,koncept.models,cv2,imquality,imquality.brisque,lpips,mixsimulator.*,networkx.*,cdt.*,pymoo,pymoo.*,bayes_optim.*,olympus.*,pymoo,pymoo.*,pybullet,pybullet_envs,pybulletgym,pyvirtualdisplay,nlopt,aquacrop.*,gomea]
 ignore_missing_imports = True
 
 [mypy-nevergrad.functions.rl.*,torchvision,torchvision.*,nevergrad.functions.games.*,nevergrad.functions.multiobjective.pyhv,nevergrad.optimization.test_doc,nevergrad.functions.gym.multigym,nevergrad.functions.gym.tuple_gym_env,nevergrad.common.sphere,nevergrad.examples.*]

nevergrad/__init__.py

@@ -15,4 +15,4 @@
 __all__ = ["optimizers", "families", "callbacks", "p", "typing", "errors", "ops"]
 
 
-__version__ = "0.14.0"
+__version__ = "1.0.0"

nevergrad/common/sphere.py

@@ -776,7 +776,7 @@ def quasi_randomize(pointset, method):
         if len(shape) > 1 and shape[0] > 5:
             x = dispersion(n, shape, conv=[int(s / 3) for s in list(shape)[:-1]])
         else:
-            x = rs_metric_all(n, shape)
+            x = rs_ng_DiagonalCMA(n, shape)
     else:
         x = get_a_point_set(n, shape, method)
     x = normalize(x)

nevergrad/optimization/base.py

@@ -90,6 +90,7 @@ def __init__(
         # you can also replace or reinitialize this random state
         self.num_workers = int(num_workers)
         self.budget = budget
+        self.optim_curve: tp.List[tp.Any] = []
 
         # How do we deal with cheap constraints i.e. constraints which are fast and use low resources and easy ?
         # True ==> we penalize them (infinite values for candidates which violate the constraint).
@@ -348,6 +349,10 @@ def tell(
             raise TypeError(
                 f'"tell" method only supports float values but the passed loss was: {loss} (type: {type(loss)}.'
             )
+        if isinstance(loss, float) and (
+            len(self.optim_curve) == 0 or self.num_tell > self.optim_curve[-1][0] * 1.1
+        ):
+            self.optim_curve += [(self.num_tell, loss)]
         # check Parameter
         if not isinstance(candidate, p.Parameter):
             raise TypeError(

nevergrad/optimization/recastlib.py

@@ -42,6 +42,9 @@ def __init__(
             method
             in [
                 "CmaFmin2",
+                "gomea",
+                "gomeablock",
+                "gomeatree",
                 "SMAC3",
                 "BFGS",
                 "LBFGSB",
@@ -362,6 +365,32 @@ def smac2_obj(p, seed: int = 0):
             #                    best_res = cost
             #                    best_x = weakself._normalizer.backward(np.asarray(x, dtype=float))
             #
+
+            elif "gomea" in weakself.method:
+                import gomea
+
+                class gomea_function(gomea.fitness.BBOFitnessFunctionRealValued):
+                    def objective_function(self, objective_index, data):  # type: ignore
+                        if weakself._normalizer is not None:
+                            data = weakself._normalizer.backward(np.asarray(data, dtype=np.float32))
+                        return objective_function(data)
+
+                gomea_f = gomea_function(weakself.dimension)
+                lm = {
+                    "gomea": gomea.linkage.Univariate(),
+                    "gomeablock": gomea.linkage.BlockMarginalProduct(2),
+                    "gomeatree": gomea.linkage.LinkageTree("NMI".encode(), True, 0),
+                }[weakself.method]
+                rvgom = gomea.RealValuedGOMEA(
+                    fitness=gomea_f,
+                    linkage_model=lm,
+                    lower_init_range=0.0,
+                    upper_init_range=1.0,
+                    max_number_of_evaluations=budget,
+                )
+                rvgom.run()
+                best_x = gomea_f.best_x
+
             elif weakself.method == "CmaFmin2":
                 import cma  # type: ignore
 
@@ -469,6 +498,9 @@ def __init__(self, *, method: str = "Nelder-Mead", random_restart: bool = False)
 BOBYQA = NonObjectOptimizer(method="BOBYQA").set_name("BOBYQA", register=True)
 NelderMead = NonObjectOptimizer(method="Nelder-Mead").set_name("NelderMead", register=True)
 CmaFmin2 = NonObjectOptimizer(method="CmaFmin2").set_name("CmaFmin2", register=True)
+GOMEA = NonObjectOptimizer(method="gomea").set_name("GOMEA", register=True)
+GOMEABlock = NonObjectOptimizer(method="gomeablock").set_name("GOMEABlock", register=True)
+GOMEATree = NonObjectOptimizer(method="gomeatree").set_name("GOMEATree", register=True)
 # NLOPT = NonObjectOptimizer(method="NLOPT").set_name("NLOPT", register=True)
 Powell = NonObjectOptimizer(method="Powell").set_name("Powell", register=True)
 RPowell = NonObjectOptimizer(method="Powell", random_restart=True).set_name("RPowell", register=True)

nevergrad/optimization/test_optimizerlib.py

@@ -144,6 +144,9 @@ def check_optimizer(
     "NLOPT_GN_CRS2_LM",
     "NLOPT_GN_ISRES",
     "NLOPT_GN_ESCH",
+    "GOMEABlock",
+    "GOMEA",
+    "GOMEATree",
 ]
 
 
@@ -299,6 +302,7 @@ def test_optimizers_minimal(name: str) -> None:
             "CMAbounded",
             "Tiny",
             "iscrete",
+            "GOMEA",
             "para",
             "SPSA",
             "EDA",
@@ -314,6 +318,7 @@ def test_optimizers_minimal(name: str) -> None:
             "Small",
             "small",
             "Chain",
+            "Tree",
             "Mix",
             "Micro",
             "Naive",

nevergrad/optimization/test_suggest.py

@@ -23,7 +23,7 @@
 
 def suggestable(name: str) -> bool:
     # Some methods are not good with suggestions.
-    keywords = ["TBPSA", "BO", "EMNA", "EDA", "BO", "Stupid", "Pymoo"]
+    keywords = ["TBPSA", "BO", "EMNA", "EDA", "BO", "Stupid", "Pymoo", "GOMEA"]
     return not any(x in name for x in keywords)
 
 
@@ -75,6 +75,7 @@ def good_at_suggest(name: str) -> bool:
         "Multi",
         "Anisotropic",
         "BSO",
+        "GOMEA",
         "Sparse",
         "Adaptive",
         "Doerr",

requirements/bench.txt

@@ -56,3 +56,4 @@ pytest-circleci-parallelized
 Py-BOBYQA>=1.2
 ax-platform
 loguru # for fcmaes
+gomea