Merge pull request #2 from jungtaekkim/0.1.5

0.1.5

.travis.yml

@@ -10,6 +10,7 @@ python:
 install:
   - pip install .
 script:
+  - pip install scipy
   - pytest
 notifications:
   slack: bayeso:FWBoHH9TMqjKUJWkZxCaTNVE

README.md

@@ -1,5 +1,5 @@
 # BayesO Benchmarks
-[![Build Status](https://travis-ci.org/jungtaekkim/bayeso-benchmarks.svg?branch=main)](https://travis-ci.org/jungtaekkim/bayeso-benchmarks)
+[![Build Status](https://travis-ci.com/jungtaekkim/bayeso-benchmarks.svg?branch=main)](https://travis-ci.com/jungtaekkim/bayeso-benchmarks)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
 Benchmarks for Bayesian optimization.
@@ -50,6 +50,20 @@ The following `requirements` files include the package list, the purpose of whic
 
 * `requirements-dev.txt`: It is for developing the `bayeso-benchmarks` package.
 
+## Simple Example
+A simple example on Branin function is shown below.
+```python
+from bayeso_benchmarks import Branin
+
+obj_fun = Branin()
+bounds = obj_fun.get_bounds()
+
+X = obj_fun.sample_uniform(100)
+
+Y = obj_fun.output(X)
+Y_noise = obj_fun.output_gaussian_noise(X)
+```
+
 ## Author
 * [Jungtaek Kim](http://jungtaek.github.io) (POSTECH)
 

bayeso_benchmarks/__init__.py

@@ -1,6 +1,32 @@
 #
 # author: Jungtaek Kim ([email protected])
-# last updated: November 5, 2020
+# last updated: June 24, 2021
 #
 
-__version__ = '0.1.4'
+__version__ = '0.1.5'
+
+
+from bayeso_benchmarks.inf_dim_ackley import Ackley
+from bayeso_benchmarks.inf_dim_cosines import Cosines
+from bayeso_benchmarks.inf_dim_rosenbrock import Rosenbrock
+from bayeso_benchmarks.inf_dim_sphere import Sphere
+
+from bayeso_benchmarks.one_dim_constant import Constant
+from bayeso_benchmarks.one_dim_gramacyandlee2012 import GramacyAndLee2012
+from bayeso_benchmarks.one_dim_linear import Linear
+from bayeso_benchmarks.one_dim_step import Step
+
+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_dejong5 import DeJong5
+from bayeso_benchmarks.two_dim_dropwave import DropWave
+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_michalewicz import Michalewicz
+from bayeso_benchmarks.two_dim_sixhumpcamel import SixHumpCamel
+from bayeso_benchmarks.two_dim_threehumpcamel import ThreeHumpCamel
+
+from bayeso_benchmarks.three_dim_hartmann3d import Hartmann3D
+from bayeso_benchmarks.six_dim_hartmann6d import Hartmann6D

bayeso_benchmarks/benchmark_base.py

@@ -1,6 +1,6 @@
 #
 # author: Jungtaek Kim ([email protected])
-# last updated: February 8, 2021
+# last updated: May 1, 2021
 #
 
 import numpy as np
@@ -126,7 +126,7 @@ def output_sparse_gaussian_noise(self, X, scale_noise=0.1, sparsity=0.01):
             num_X = X.shape[0]
         else:
             num_X = 1
-            
+
         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)
@@ -138,6 +138,44 @@ def output_sparse_gaussian_noise(self, X, scale_noise=0.1, sparsity=0.01):
         assert Y.shape[1] == 1
         return Y
 
+    def output_student_t_noise(self, X, scale_noise=0.01, dof=4.0):
+        assert isinstance(scale_noise, float)
+        assert isinstance(dof, float)
+
+        by = self._output(X)
+        by += scale_noise * self.random_state.standard_t(dof, size=by.shape[0])
+
+        Y = np.expand_dims(by, axis=1)
+
+        assert len(Y.shape) == 2
+        assert Y.shape[1] == 1
+        return Y
+
+    def output_sparse_student_t_noise(self, X, scale_noise=0.1, dof=4.0, sparsity=0.01):
+        assert isinstance(scale_noise, float)
+        assert isinstance(dof, float)
+        assert isinstance(sparsity, float)
+        assert sparsity >= 0.0 and sparsity <= 1.0
+        assert sparsity < 0.5
+
+        by = self._output(X)
+
+        if len(X.shape) == 2:
+            num_X = X.shape[0]
+        else:
+            num_X = 1
+
+        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)
+        by += scale_noise * noise
+
+        Y = np.expand_dims(by, axis=1)
+
+        assert len(Y.shape) == 2
+        assert Y.shape[1] == 1
+        return Y
+
     def validate_properties(self):
         shape_bounds = self.get_bounds().shape
 
@@ -154,7 +192,7 @@ def validate_properties(self):
         else:
             assert self.dimensionality == shape_bounds[0] == shape_global_minimizers[1]
 
-    def get_grids(self, num_grids):
+    def sample_grids(self, num_grids):
         assert isinstance(num_grids, int)
 
         list_grids = []
@@ -167,3 +205,21 @@ def get_grids(self, num_grids):
         grids = np.vstack(tuple(list_grids))
         grids = grids.T
         return grids
+
+    def sample_uniform(self, num_points, seed=None):
+        assert isinstance(num_points, int)
+        assert isinstance(seed, (type(None), int))
+
+        random_state_ = np.random.RandomState(seed)
+
+        if self.dimensionality is np.inf:
+            dim_problem = self.dim_problem
+        else:
+            dim_problem = self.dimensionality
+
+        bounds = self.get_bounds()
+
+        points = random_state_.uniform(size=(num_points, dim_problem))
+        points = (bounds[:, 0] + (bounds[:, 1] - bounds[:, 0])) * points
+
+        return points

bayeso_benchmarks/inf_dim_ackley.py

@@ -24,8 +24,9 @@ def fun_target(bx, dim_bx,
 
 
 class Ackley(Function):
-    def __init__(self, dim_problem):
+    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([
@@ -39,4 +40,4 @@ def __init__(self, dim_problem):
 
         function = lambda bx: fun_target(bx, dim_problem)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/inf_dim_cosines.py

@@ -17,8 +17,9 @@ def fun_target(bx, dim_bx):
 
 
 class Cosines(Function):
-    def __init__(self, dim_problem):
+    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([
@@ -32,4 +33,4 @@ def __init__(self, dim_problem):
 
         function = lambda bx: fun_target(bx, dim_problem)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/inf_dim_rosenbrock.py

@@ -20,8 +20,9 @@ def fun_target(bx, dim_bx):
 
 
 class Rosenbrock(Function):
-    def __init__(self, dim_problem):
+    def __init__(self, dim_problem, seed=None):
         assert isinstance(dim_problem, int)
+        assert isinstance(seed, (type(None), int))
         assert dim_problem > 1
 
         dim_bx = np.inf
@@ -36,4 +37,4 @@ def __init__(self, dim_problem):
 
         function = lambda bx: fun_target(bx, dim_problem)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/inf_dim_sphere.py

@@ -20,8 +20,9 @@ def fun_target(bx, dim_bx):
 
 
 class Sphere(Function):
-    def __init__(self, dim_problem):
+    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([
@@ -35,4 +36,4 @@ def __init__(self, dim_problem):
 
         function = lambda bx: fun_target(bx, dim_problem)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, dim_problem=dim_problem, seed=seed)

bayeso_benchmarks/one_dim_constant.py

@@ -22,10 +22,12 @@ def __init__(self,
         bounds = np.array([
             [-10.0, 10.0],
         ]),
-        constant=0.0
+        constant=0.0,
+        seed=None
     ):
         assert isinstance(constant, float)
         assert isinstance(bounds, np.ndarray)
+        assert isinstance(seed, (type(None), int))
         assert len(bounds.shape) == 2
         assert bounds.shape[0] == 1
         assert bounds.shape[1] == 2
@@ -41,4 +43,4 @@ def __init__(self,
         global_minimum = constant
         function = lambda bx: fun_target(bx, dim_bx, constant)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/one_dim_gramacyandlee2012.py

@@ -17,7 +17,9 @@ def fun_target(bx, dim_bx):
 
 
 class GramacyAndLee2012(Function):
-    def __init__(self):
+    def __init__(self, seed=None):
+        assert isinstance(seed, (type(None), int))
+
         dim_bx = 1
         bounds = np.array([
             [0.5, 2.5],
@@ -28,4 +30,4 @@ def __init__(self):
         global_minimum = -0.86901113
         function = lambda bx: fun_target(bx, dim_bx)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/one_dim_linear.py

@@ -22,10 +22,12 @@ def __init__(self,
         bounds=np.array([
             [-10, 10],
         ]),
-        slope=1.0
+        slope=1.0,
+        seed=None
     ):
         assert isinstance(slope, float)
         assert isinstance(bounds, np.ndarray)
+        assert isinstance(seed, (type(None), int))
         assert len(bounds.shape) == 2
         assert bounds.shape[0] == 1
         assert bounds.shape[1] == 2
@@ -45,4 +47,4 @@ def __init__(self,
             global_minimum = slope * bounds[0, 1]
         function = lambda bx: fun_target(bx, dim_bx, slope)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/one_dim_step.py

@@ -35,9 +35,11 @@ class Step(Function):
     def __init__(self,
         steps=[-10., -5., 0., 5., 10.],
         step_values=[-2., 0., 1., -1.],
+        seed=None
     ):
         assert isinstance(steps, list)
         assert isinstance(step_values, list)
+        assert isinstance(seed, (type(None), int))
         assert len(steps) == len(step_values) + 1
         assert isinstance(steps[0], float)
         assert isinstance(step_values[0], float)
@@ -53,4 +55,4 @@ def __init__(self,
         global_minimum = np.min(step_values)
         function = lambda bx: fun_target(bx, dim_bx, steps, step_values)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/six_dim_hartmann6d.py

@@ -46,9 +46,11 @@ def __init__(self,
             [0.0, 1.0],
             [0.0, 1.0],
             [0.0, 1.0],
-        ])
+        ]),
+        seed=None
     ):
         assert isinstance(bounds, np.ndarray)
+        assert isinstance(seed, (type(None), int))
         assert len(bounds.shape) == 2
         assert bounds.shape[1] == 2
 
@@ -61,4 +63,4 @@ def __init__(self,
         global_minimum = -3.322368
         function = lambda bx: fun_target(bx, dim_bx)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/three_dim_hartmann3d.py

@@ -43,9 +43,11 @@ def __init__(self,
             [0.0, 1.0],
             [0.0, 1.0],
             [0.0, 1.0],
-        ])
+        ]),
+        seed=None
     ):
         assert isinstance(bounds, np.ndarray)
+        assert isinstance(seed, (type(None), int))
         assert len(bounds.shape) == 2
         assert bounds.shape[1] == 2
 
@@ -58,4 +60,4 @@ def __init__(self,
         global_minimum = -3.86278
         function = lambda bx: fun_target(bx, dim_bx)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/two_dim_beale.py

@@ -17,7 +17,9 @@ def fun_target(bx, dim_bx):
 
 
 class Beale(Function):
-    def __init__(self):
+    def __init__(self, seed=None):
+        assert isinstance(seed, (type(None), int))
+
         dim_bx = 2
         bounds = np.array([
             [-4.5, 4.5],
@@ -29,4 +31,4 @@ def __init__(self):
         global_minimum = 0.0
         function = lambda bx: fun_target(bx, dim_bx)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)

bayeso_benchmarks/two_dim_bohachevsky.py

@@ -17,7 +17,9 @@ def fun_target(bx, dim_bx):
 
 
 class Bohachevsky(Function):
-    def __init__(self):
+    def __init__(self, seed=None):
+        assert isinstance(seed, (type(None), int))
+
         dim_bx = 2
         bounds = np.array([
             [-100.0, 100.0],
@@ -29,4 +31,4 @@ def __init__(self):
         global_minimum = 0.0
         function = lambda bx: fun_target(bx, dim_bx)
 
-        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function)
+        Function.__init__(self, dim_bx, bounds, global_minimizers, global_minimum, function, seed=seed)