Disable GPy for newer numpy

GPy can not be used for newer numpy

HEBO/hebo/models/__init__.py

@@ -7,4 +7,10 @@
 # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
 # PARTICULAR PURPOSE. See the MIT License for more details.
 
+has_gpy = True
+try:
+    from numpy.testing import Tester
+except ImportError:
+    has_gpy = False
+
 from . import util, scalers, model_factory, layers, base_model

HEBO/hebo/models/gp/__init__.py

@@ -7,4 +7,10 @@
 # WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
 # PARTICULAR PURPOSE. See the MIT License for more details.
 
-from . import gp, gpy_wgp, gpy_mlp
+
+from hebo.models import has_gpy
+
+if has_gpy:
+    from . import gp, gpy_wgp, gpy_mlp
+else:
+    from . import gp

HEBO/hebo/models/gp/gpy_mlp.py

@@ -21,6 +21,10 @@
 from ..scalers import TorchMinMaxScaler, TorchStandardScaler
 from ..util import filter_nan
 
+from hebo.models import has_gpy
+if not has_gpy:
+    raise ImportError('GPy can not be installed with current numpy version, use GP instead')
+
 
 class GPyMLPGP(BaseModel):
     """

HEBO/hebo/models/gp/gpy_wgp.py

@@ -14,6 +14,9 @@
 from ..layers import EmbTransform, OneHotTransform
 from ..scalers import TorchMinMaxScaler, TorchStandardScaler
 from ..util import filter_nan
+from hebo.models import has_gpy
+if not has_gpy:
+    raise ImportError('GPy can not be installed with current numpy version, use GP instead')
 
 import GPy
 import torch

HEBO/hebo/models/model_factory.py

@@ -8,12 +8,14 @@
 # PARTICULAR PURPOSE. See the MIT License for more details.
 
 import torch
+from hebo.models import has_gpy
 from .base_model import BaseModel
 from .gp.svgp import SVGP
 from .gp.svidkl import SVIDKL
 from .gp.gp import GP
-from .gp.gpy_wgp import GPyGP
-from .gp.gpy_mlp import GPyMLPGP
+if has_gpy:
+    from .gp.gpy_wgp import GPyGP
+    from .gp.gpy_mlp import GPyMLPGP
 from .rf.rf import RF
 
 from .nn.mcbn import MCBNEnsemble
@@ -29,20 +31,34 @@
 except ImportError:
     has_catboost = False
 
-model_dict = {
-        'svidkl'  : SVIDKL,
-        'svgp'  : SVGP,
-        'gp'  : GP,
-        'gpy' : GPyGP,
-        'gpy_mlp' : GPyMLPGP, 
-        'rf'  : RF,
-        'deep_ensemble' : DeepEnsemble,
-        'psgld' : pSGLDEnsemble,
-        'mcbn' : MCBNEnsemble, 
-        'masked_deep_ensemble' : MaskedDeepEnsemble,
-        'fe_deep_ensemble': FeDeepEnsemble, 
-        'gumbel': GumbelDeepEnsemble, 
-        }
+if has_gpy:
+    model_dict = {
+            'svidkl'  : SVIDKL,
+            'svgp'  : SVGP,
+            'gp'  : GP,
+            'gpy' : GPyGP,
+            'gpy_mlp' : GPyMLPGP, 
+            'rf'  : RF,
+            'deep_ensemble' : DeepEnsemble,
+            'psgld' : pSGLDEnsemble,
+            'mcbn' : MCBNEnsemble, 
+            'masked_deep_ensemble' : MaskedDeepEnsemble,
+            'fe_deep_ensemble': FeDeepEnsemble, 
+            'gumbel': GumbelDeepEnsemble, 
+            }
+else:
+    model_dict = {
+            'svidkl'  : SVIDKL,
+            'svgp'  : SVGP,
+            'gp'  : GP,
+            'rf'  : RF,
+            'deep_ensemble' : DeepEnsemble,
+            'psgld' : pSGLDEnsemble,
+            'mcbn' : MCBNEnsemble, 
+            'masked_deep_ensemble' : MaskedDeepEnsemble,
+            'fe_deep_ensemble': FeDeepEnsemble, 
+            'gumbel': GumbelDeepEnsemble, 
+            }
 if has_catboost:
     model_dict['catboost'] = CatBoost
 
@@ -75,7 +91,7 @@ def __init__(self,
         self.model_name = self.conf.get('base_model_name', 'gp')
         self.model_conf = {k : v for k, v in self.conf.items() if k != 'model_name'}
         self.models     = [model_dict[self.model_name](num_cont, num_enum, 1, **self.model_conf) for _ in range(num_out)]
-    
+
 
     def fit(self, Xc, Xe, y):
         for i in range(self.num_out):

HEBO/hebo/optimizers/bo.py

@@ -11,20 +11,23 @@
 import pandas as pd
 import torch
 
+from hebo.models import has_gpy
 from hebo.design_space.design_space import DesignSpace
 from hebo.models.model_factory import get_model
 from hebo.acquisitions.acq import LCB
 from hebo.acq_optimizers.evolution_optimizer import EvolutionOpt
 
 from .abstract_optimizer import AbstractOptimizer
 
+default_gp = 'gpy' if has_gpy else 'gp'
+
 class BO(AbstractOptimizer):
     support_combinatorial = True
     support_contextual    = True
     def __init__(
             self,
             space : DesignSpace,
-            model_name  = 'gpy',
+            model_name  = default_gp,
             rand_sample = None,
             acq_cls     = None, 
             acq_conf    = None):

HEBO/hebo/optimizers/hebo.py

@@ -18,6 +18,7 @@
 from torch.quasirandom import SobolEngine
 from sklearn.preprocessing import power_transform
 
+from hebo.models import has_gpy
 from hebo.design_space.design_space import DesignSpace
 from hebo.models.model_factory import get_model
 from hebo.acquisitions.acq import MACE, Mean, Sigma
@@ -27,11 +28,13 @@
 
 torch.set_num_threads(min(1, torch.get_num_threads()))
 
+default_gp = 'gpy' if has_gpy else 'gp'
+
 class HEBO(AbstractOptimizer):
     support_parallel_opt  = True
     support_combinatorial = True
     support_contextual    = True
-    def __init__(self, space, model_name = 'gpy', rand_sample = None, acq_cls = MACE, es = 'nsga2', model_config = None,
+    def __init__(self, space, model_name = default_gp, rand_sample = None, acq_cls = MACE, es = 'nsga2', model_config = None,
                  scramble_seed: Optional[int] = None ):
         """
         model_name  : surrogate model to be used

HEBO/hebo/optimizers/hebo_contextual.py

@@ -16,14 +16,17 @@
 from .hebo import HEBO
 from hebo.acquisitions.acq import Acquisition, MACE
 
+from hebo.models import has_gpy
+default_gp = 'gpy' if has_gpy else 'gp'
+
 class HEBO_VectorContextual(AbstractOptimizer):
     support_parallel_opt  = True
     support_combinatorial = True
     support_contextual    = True
     def __init__(self, 
             space,
             context_dict : dict, 
-            model_name   : str = 'gpy',
+            model_name   : str = default_gp,
             rand_sample  : int = None
             ):
         self.hebo         = HEBO(space, model_name, rand_sample)

HEBO/hebo/optimizers/hebo_embedding.py

@@ -23,6 +23,9 @@
 from .abstract_optimizer import AbstractOptimizer
 from .hebo import HEBO
 
+from hebo.models import has_gpy
+default_gp = 'gpy' if has_gpy else 'gp'
+
 torch.set_num_threads(min(1, torch.get_num_threads()))
 
 def gen_emb_space(eff_dim : int, scale : float) -> DesignSpace:
@@ -86,7 +89,7 @@ class HEBO_Embedding(AbstractOptimizer):
     support_contextual    = False
     def __init__(self,
             space : DesignSpace, 
-            model_name      = 'gpy',
+            model_name      = default_gp,
             eff_dim : int   = 1,
             scale   : float = 1,
             strategy : str  = 'alebo',

HEBO/hebo/optimizers/noisy_opt.py

@@ -15,6 +15,7 @@
 from copy import deepcopy
 from torch.quasirandom import SobolEngine
 
+from hebo.models import has_gpy
 from hebo.design_space.design_space import DesignSpace
 from hebo.models.model_factory import get_model
 from hebo.acquisitions.acq import NoisyAcq, Mean, Sigma
@@ -23,11 +24,13 @@
 
 torch.set_num_threads(min(1, torch.get_num_threads()))
 
+default_gp = 'gpy' if has_gpy else 'gp'
+
 class NoisyOpt(HEBO):
     support_parallel_opt  = True
     support_combinatorial = True
     support_contextual    = True
-    def __init__(self, space, model_name = 'gpy', rand_sample = None, es = 'nsga2', model_config = None):
+    def __init__(self, space, model_name = default_gp, rand_sample = None, es = 'nsga2', model_config = None):
         """
         model_name : surrogate model to be used
         rand_iter  : iterations to perform random sampling

HEBO/test/test_optimizer.py

@@ -18,7 +18,6 @@
 from hebo.optimizers.noisy_opt import NoisyOpt
 from hebo.optimizers.hebo import HEBO
 from hebo.optimizers.bo import BO
-from hebo.optimizers.vcbo import VCBO
 from hebo.optimizers.general import GeneralBO
 from hebo.optimizers.evolution import Evolution
 from hebo.optimizers.hebo_contextual import HEBO_VectorContextual
@@ -35,7 +34,7 @@
 def obj(x : pd.DataFrame) -> np.ndarray:
     return x['x0'].values.astype(float).reshape(-1, 1) ** 2
 
-@pytest.mark.parametrize('model_name', ['gp', 'gpy', 'rf', 'deep_ensemble', 'gpy_mlp']) 
+@pytest.mark.parametrize('model_name', ['gp', 'rf', 'deep_ensemble']) 
 @pytest.mark.parametrize('opt_cls', [BO, HEBO, GeneralBO, NoisyOpt, Evolution], ids = ['bo', 'hebo', 'general', 'noisy', 'evolution'])
 def test_opt(model_name, opt_cls):
     space = DesignSpace().parse([
@@ -55,18 +54,6 @@ def test_opt(model_name, opt_cls):
         if num_suggest > 11:
             break
 
-def test_vcbo():
-    space = DesignSpace().parse([
-        {'name' : 'x0', 'type' : 'num', 'lb' : -3, 'ub' : 7},
-        ])
-    opt = VCBO(space, rand_sample = 8)
-    num_suggest = 0
-    for i in range(11):
-        num_suggest = 8 if opt.support_parallel_opt else 1
-        rec = opt.suggest(n_suggestions = num_suggest)
-        num_suggest += rec.shape[0]
-        if num_suggest > 11:
-            break
 
 @pytest.mark.parametrize('start_from_mu', [True, False])
 def test_cmaes(start_from_mu):