Merge pull request #4 from jungtaekkim/0.1.7

0.1.7

.github/workflows/pytest.yml

@@ -12,7 +12,8 @@ jobs:
           - '3.8'
           - '3.9'
           - '3.10'
-    runs-on: ubuntu-latest
+          - '3.11'
+    runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v2
       - uses: actions/setup-python@v2

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2019-2022 Jungtaek Kim
+Copyright (c) 2019-2023 Jungtaek Kim
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -1,10 +1,16 @@
-# BayesO Benchmarks
+<p align="center">
+<img src="https://github.com/jungtaekkim/bayeso/blob/main/docs/_static/assets/logo_bayeso_capitalized.svg" width="400" />
+</p>
+
+# BayesO Benchmarks: Benchmark Functions for Bayesian Optimization
 [![Build Status](https://github.com/jungtaekkim/bayeso-benchmarks/actions/workflows/pytest.yml/badge.svg)](https://github.com/jungtaekkim/bayeso-benchmarks/actions/workflows/pytest.yml)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
-Benchmarks for Bayesian optimization.
+This repository provides the implementation of benchmark functions for Bayesian optimization.
 The details of benchmark functions can be found in [these notes](https://jungtaek.github.io/notes/benchmarks_bo.pdf).
 
+* [https://bayeso.org](https://bayeso.org)
+
 ## Installation
 We recommend installing it with `virtualenv`.
 You can choose one of three installation options.
@@ -64,8 +70,5 @@ Y = obj_fun.output(X)
 Y_noise = obj_fun.output_gaussian_noise(X)
 ```
 
-## Author
-* [Jungtaek Kim](https://jungtaek.github.io) (POSTECH)
-
 ## License
 [MIT License](LICENSE)

bayeso_benchmarks/__init__.py

@@ -1,16 +1,20 @@
 #
 # author: Jungtaek Kim (jtkim@postech.ac.kr)
-# last updated: October 23, 2021
+# last updated: January 6, 2023
 #
 
 
-__version__ = '0.1.6'
+__version__ = '0.1.7'
 
 
 from bayeso_benchmarks.inf_dim_ackley import Ackley
 from bayeso_benchmarks.inf_dim_cosines import Cosines
+from bayeso_benchmarks.inf_dim_griewank import Griewank
+from bayeso_benchmarks.inf_dim_levy import Levy
+from bayeso_benchmarks.inf_dim_rastrigin import Rastrigin
 from bayeso_benchmarks.inf_dim_rosenbrock import Rosenbrock
 from bayeso_benchmarks.inf_dim_sphere import Sphere
+from bayeso_benchmarks.inf_dim_zakharov import Zakharov
 
 from bayeso_benchmarks.one_dim_constant import Constant
 from bayeso_benchmarks.one_dim_gramacyandlee2012 import GramacyAndLee2012
@@ -20,17 +24,58 @@
 from bayeso_benchmarks.two_dim_beale import Beale
 from bayeso_benchmarks.two_dim_bohachevsky import Bohachevsky
 from bayeso_benchmarks.two_dim_branin import Branin
+from bayeso_benchmarks.two_dim_bukin6 import Bukin6
 from bayeso_benchmarks.two_dim_dejong5 import DeJong5
 from bayeso_benchmarks.two_dim_dropwave import DropWave
+from bayeso_benchmarks.two_dim_easom import Easom
 from bayeso_benchmarks.two_dim_eggholder import Eggholder
 from bayeso_benchmarks.two_dim_goldsteinprice import GoldsteinPrice
 from bayeso_benchmarks.two_dim_holdertable import HolderTable
 from bayeso_benchmarks.two_dim_kim1 import Kim1
 from bayeso_benchmarks.two_dim_kim2 import Kim2
 from bayeso_benchmarks.two_dim_kim3 import Kim3
 from bayeso_benchmarks.two_dim_michalewicz import Michalewicz
+from bayeso_benchmarks.two_dim_shubert import Shubert
 from bayeso_benchmarks.two_dim_sixhumpcamel import SixHumpCamel
 from bayeso_benchmarks.two_dim_threehumpcamel import ThreeHumpCamel
 
+from bayeso_benchmarks.four_dim_colville import Colville
 from bayeso_benchmarks.three_dim_hartmann3d import Hartmann3D
 from bayeso_benchmarks.six_dim_hartmann6d import Hartmann6D
+
+
+all_benchmarks = [
+    Ackley,
+    Cosines,
+    Griewank,
+    Levy,
+    Rastrigin,
+    Rosenbrock,
+    Sphere,
+    Zakharov,
+    Constant,
+    GramacyAndLee2012,
+    Linear,
+    Step,
+    Beale,
+    Bohachevsky,
+    Branin,
+    Bukin6,
+    DeJong5,
+    DropWave,
+    Easom,
+    Eggholder,
+    GoldsteinPrice,
+    HolderTable,
+    Kim1,
+    Kim2,
+    Kim3,
+    Michalewicz,
+    Shubert,
+    SixHumpCamel,
+    ThreeHumpCamel,
+    Colville,
+    Hartmann3D,
+    Hartmann6D,
+]
+num_benchmarks = len(all_benchmarks)

bayeso_benchmarks/benchmark_base.py

@@ -1,21 +1,22 @@
 #
 # author: Jungtaek Kim (jtkim@postech.ac.kr)
-# last updated: May 1, 2021
+# last updated: December 13, 2022
 #
 
 import numpy as np
 
 EPSILON = 1e-4
 
 
-class Function(object):
+class Function:
     def __init__(self, dimensionality, bounds, global_minimizers, global_minimum, function, dim_problem=None, seed=None):
         assert isinstance(dimensionality, int) or dimensionality is np.inf
         assert isinstance(bounds, np.ndarray)
         assert isinstance(global_minimizers, np.ndarray)
         assert isinstance(global_minimum, float)
         assert callable(function)
-        assert isinstance(dim_problem, int) or dim_problem is None
+        assert isinstance(dim_problem, (type(None), int))
+        assert isinstance(seed, (type(None), int))
         assert len(bounds.shape) == 2
         assert bounds.shape[1] == 2
         assert (bounds[:, 0] <= bounds[:, 1]).all()
@@ -30,6 +31,7 @@ def __init__(self, dimensionality, bounds, global_minimizers, global_minimum, fu
         self.random_state = np.random.RandomState(seed)
 
         self.validate_properties()
+        self.set_name()
 
     @property
     def dimensionality(self):
@@ -47,6 +49,14 @@ def global_minimizers(self):
     def global_minimum(self):
         return self._global_minimum
 
+    def set_name(self):
+        name = self.__class__.__name__.lower()
+
+        if self.dimensionality is np.inf:
+            self.name = f'{name}_{self.dim_problem}'
+        else:
+            self.name = name
+
     def get_bounds(self):
         if self.dimensionality is np.inf:
             return np.array(list(self.bounds) * self.dim_problem)
@@ -134,7 +144,7 @@ def output_sparse_gaussian_noise(self, X, scale_noise=0.1, sparsity=0.01):
 
         noise = self.random_state.randn(num_X)
         mask = self.random_state.uniform(low=0.0, high=1.0, size=num_X) < sparsity
-        noise *= mask.astype(np.float)
+        noise *= mask.astype(float)
         by += scale_noise * noise
 
         Y = np.expand_dims(by, axis=1)
@@ -172,7 +182,7 @@ def output_sparse_student_t_noise(self, X, scale_noise=0.1, dof=4.0, sparsity=0.
 
         noise = self.random_state.standard_t(dof, size=num_X)
         mask = self.random_state.uniform(low=0.0, high=1.0, size=num_X) < sparsity
-        noise *= mask.astype(np.float)
+        noise *= mask.astype(float)
         by += scale_noise * noise
 
         Y = np.expand_dims(by, axis=1)
@@ -228,3 +238,6 @@ def sample_uniform(self, num_points, seed=None):
         points = bounds[:, 0] + (bounds[:, 1] - bounds[:, 0]) * points
 
         return points
+
+    def __call__(self, X):
+        return self.output(X)

bayeso_benchmarks/four_dim_colville.py

@@ -0,0 +1,48 @@
+#
+# author: Jungtaek Kim (jtkim@postech.ac.kr)
+# last updated: December 30, 2022
+#
+
+import numpy as np
+
+from bayeso_benchmarks.benchmark_base import Function
+
+
+def fun_target(bx, dim_bx):
+    assert len(bx.shape) == 1
+    assert bx.shape[0] == dim_bx
+
+    y = 100.0 * (bx[0]**2 - bx[1])**2
+    y += (bx[0] - 1.0)**2
+    y += (bx[2] - 1.0)**2
+    y += 90.0 * (bx[2]**2 - bx[3])**2
+    y += 10.1 * ((bx[1] - 1.0)**2 + (bx[3] - 1.0)**2)
+    y += 19.8 * (bx[1] - 1.0) * (bx[3] - 1.0)
+
+    return y
+
+
+class Colville(Function):
+    def __init__(self, seed=None):
+        assert isinstance(seed, (type(None), int))
+
+        dim_bx = 4
+        bounds = np.array([
+            [-10.0, 10.0],
+            [-10.0, 10.0],
+            [-10.0, 10.0],
+            [-10.0, 10.0],
+        ])
+        assert bounds.shape[0] == dim_bx
+        assert bounds.shape[1] == 2
+
+        global_minimizers = np.array([
+            [1.0, 1.0, 1.0, 1.0],
+        ])
+        global_minimum = 0.0
+        function = lambda bx: fun_target(bx, dim_bx)
+
+        try:
+            super().__init__(dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)
+        except:
+            super(Colville, self).__init__(dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/inf_dim_griewank.py

@@ -0,0 +1,45 @@
+#
+# author: Jungtaek Kim (jtkim@postech.ac.kr)
+# last updated: January 6, 2023
+#
+
+import numpy as np
+
+from bayeso_benchmarks.benchmark_base import Function
+
+
+def fun_target(bx, dim_bx):
+    assert len(bx.shape) == 1
+    assert bx.shape[0] == dim_bx
+
+    first_term = np.sum(bx**2 / 4000.0)
+
+    second_term = 1.0
+    for ind in range(1, dim_bx + 1):
+        second_term *= np.cos(bx[ind - 1] / np.sqrt(ind))
+
+    y = first_term - second_term + 1.0
+    return y
+
+
+class Griewank(Function):
+    def __init__(self, dim_problem, seed=None):
+        assert isinstance(dim_problem, int)
+        assert isinstance(seed, (type(None), int))
+
+        dim_bx = np.inf
+        bounds = np.array([
+            [-600.0, 600.0],
+        ])
+        global_minimizers = np.array([
+            [0.0],
+        ])
+        global_minimum = 0.0
+        dim_problem = dim_problem
+
+        function = lambda bx: fun_target(bx, dim_problem)
+
+        try:
+            super().__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)
+        except:
+            super(Griewank, self).__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/inf_dim_levy.py

@@ -0,0 +1,51 @@
+#
+# author: Jungtaek Kim (jtkim@postech.ac.kr)
+# last updated: December 29, 2022
+#
+
+import numpy as np
+
+from bayeso_benchmarks.benchmark_base import Function
+
+
+def fun_target(bx, dim_bx,
+):
+    assert len(bx.shape) == 1
+    assert bx.shape[0] == dim_bx
+
+    bw = []
+    for x in bx:
+        w = 1.0 + (x - 1.0) / 4.0
+        bw.append(w)
+    bw = np.array(bw)
+
+    y = np.sin(np.pi * bw[0])**2
+
+    for w in bw[:-1]:
+        y += (w - 1.0)**2 * (1.0 + 10.0 * np.sin(np.pi * w + 1.0)**2)
+
+    y += (bw[-1] - 1.0)**2 * (1.0 + np.sin(2.0 * np.pi * bw[-1])**2)
+    return y
+
+
+class Levy(Function):
+    def __init__(self, dim_problem, seed=None):
+        assert isinstance(dim_problem, int)
+        assert isinstance(seed, (type(None), int))
+
+        dim_bx = np.inf
+        bounds = np.array([
+            [-10.0, 10.0],
+        ])
+        global_minimizers = np.array([
+            [1.0],
+        ])
+        global_minimum = 0.0
+        dim_problem = dim_problem
+
+        function = lambda bx: fun_target(bx, dim_problem)
+
+        try:
+            super().__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)
+        except:
+            super(Levy, self).__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/inf_dim_rastrigin.py

@@ -0,0 +1,43 @@
+#
+# author: Jungtaek Kim (jtkim@postech.ac.kr)
+# last updated: December 20, 2022
+#
+
+import numpy as np
+
+from bayeso_benchmarks.benchmark_base import Function
+
+
+def fun_target(bx, dim_bx):
+    assert len(bx.shape) == 1
+    assert bx.shape[0] == dim_bx
+
+    y = 10.0 * dim_bx
+
+    for ind in range(0, dim_bx):
+        y += bx[ind]**2 - 10.0 * np.cos(2.0 * np.pi * bx[ind])
+
+    return y
+
+
+class Rastrigin(Function):
+    def __init__(self, dim_problem, seed=None):
+        assert isinstance(dim_problem, int)
+        assert isinstance(seed, (type(None), int))
+
+        dim_bx = np.inf
+        bounds = np.array([
+            [-5.12, 5.12],
+        ])
+        global_minimizers = np.array([
+            [0.0],
+        ])
+        global_minimum = 0.0
+        dim_problem = dim_problem
+
+        function = lambda bx: fun_target(bx, dim_problem)
+
+        try:
+            super().__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)
+        except:
+            super(Rastrigin, self).__init__(dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)