Improve RBF model

README.md

@@ -4,7 +4,31 @@ Surrogate models and active learning for scientific applications.
 
 ## Building
 
-This project uses [pdm](https://pdm-project.org/en/stable/) as its package manager. With pdm installed, run `pdm install` at the root of this repository to install the dependencies. The file [pyproject.toml](pyproject.toml) has the list of dependencies and configurations for the project. Use `pdm build` to build the project. The build artifacts will be in the `dist` directory. Please, find more information about pdm in its website.
+### Binaries
+
+The binaries for the latest version are available at https://github.com/NREL/bbopt/releases/latest. They can be installed through standard installation, e.g.,
+
+using pip (https://pip.pypa.io/en/stable/cli/pip_install/):
+
+```sh
+python -m pip install
+https://github.com/NREL/bbopt/archive/refs/tags/v0.4.2.tar.gz
+```
+
+using conda (https://docs.anaconda.com/working-with-conda/packages/install-packages/):
+
+```sh
+conda install
+https://github.com/NREL/bbopt/archive/refs/tags/v0.4.2.tar.gz
+```
+
+### From source
+
+This package contains a [pyproject.toml](pyproject.toml) with the list of requirements and dependencies (More about `pyproject.toml` at https://packaging.python.org/en/latest/specifications/pyproject-toml/). This project is configured to use the package manager [pdm](https://pdm-project.org/en/stable/). With pdm installed, run `pdm install` at the root of this repository to install the dependencies. The file [pyproject.toml](pyproject.toml) has the list of dependencies and configurations for the project. Use `pdm build` to build the packages binaries in the `dist` directory. Then follow the steps in the [Binaries section](#binaries) to install.
+
+### For developers
+
+Install the package manager [pdm](https://pdm-project.org/en/stable/). To install all the dependencies listted in the [pyproject.toml](pyproject.toml), run `pdm install`.
 
 ## Documentation
 

blackboxopt/__init__.py

@@ -16,7 +16,7 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 __all__ = ["acquisition", "optimize", "sampling", "rbf"]
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 
 from . import acquisition
 from . import optimize

blackboxopt/acquisition.py

@@ -31,7 +31,7 @@
     "Haoyu Jia",
     "Weslley S. Pereira",
 ]
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 __deprecated__ = False
 
 import numpy as np
@@ -53,7 +53,7 @@
 
 # Local imports
 from .sampling import NormalSampler, Sampler
-from .rbf import RbfModel, RbfType
+from .rbf import RbfModel, RbfKernel
 from .problem import (
     ProblemWithConstraint,
     ProblemNoConstraint,
@@ -376,13 +376,13 @@ def acquire(
         # Evaluate candidates
         if not listOfSurrogates:
             samples = surrogateModel.samples()
-            fx, _ = surrogateModel.eval(x)
+            fx, _ = surrogateModel(x)
         else:
             samples = surrogateModel[0].samples()
             objdim = len(surrogateModel)
             fx = np.empty((nCand, objdim))
             for i in range(objdim):
-                fx[:, i], _ = surrogateModel[i].eval(x)
+                fx[:, i], _ = surrogateModel[i](x)
 
         # Create scaled x and scaled distx
         xlow = np.array([bounds[i][0] for i in range(dim)])
@@ -494,13 +494,13 @@ def acquire(
         # Evaluate candidates
         if not listOfSurrogates:
             samples = surrogateModel.samples()
-            fx, _ = surrogateModel.eval(x)
+            fx, _ = surrogateModel(x)
         else:
             samples = surrogateModel[0].samples()
             objdim = len(surrogateModel)
             fx = np.empty((nCand, objdim))
             for i in range(objdim):
-                fx[:, i], _ = surrogateModel[i].eval(x)
+                fx[:, i], _ = surrogateModel[i](x)
 
         # Create scaled x and scaled distx
         xlow = np.array([bounds[i][0] for i in range(dim)])
@@ -601,7 +601,7 @@ def acquire(
                 else np.random.choice(self.cycleLength + 2)
             )
             if sample_stage == 0:  # InfStep - minimize Mu_n
-                LDLt = ldl(surrogateModel.get_RBFmatrix())
+                LDLt = ldl(surrogateModel.get_RBFmatrix(smoothing=0))
                 problem = ProblemWithConstraint(
                     lambda x: surrogateModel.mu_measure(x, LDLt=LDLt),
                     lambda x: self.tol
@@ -636,7 +636,7 @@ def acquire(
                 assert len(fbounds) == 2
                 # find min of surrogate model
                 problem = ProblemNoConstraint(
-                    lambda x: surrogateModel.eval(x)[0],
+                    lambda x: surrogateModel(x)[0],
                     bounds,
                     surrogateModel.iindex,
                 )
@@ -657,7 +657,7 @@ def acquire(
                 )  # target for objective function value
 
                 # use GA method to minimize bumpiness measure
-                LDLt = ldl(surrogateModel.get_RBFmatrix())
+                LDLt = ldl(surrogateModel.get_RBFmatrix(smoothing=0))
                 problem = ProblemWithConstraint(
                     lambda x: surrogateModel.bumpiness_measure(
                         x, f_target, LDLt
@@ -691,7 +691,7 @@ def acquire(
                 assert len(fbounds) == 2
                 # find the minimum of RBF surface
                 problem = ProblemNoConstraint(
-                    lambda x: surrogateModel.eval(x)[0],
+                    lambda x: surrogateModel(x)[0],
                     bounds,
                     surrogateModel.iindex,
                 )
@@ -723,7 +723,7 @@ def acquire(
                         fbounds[0]
                     )  # target value
                     # use GA method to minimize bumpiness measure
-                    LDLt = ldl(surrogateModel.get_RBFmatrix())
+                    LDLt = ldl(surrogateModel.get_RBFmatrix(smoothing=0))
                     problem = ProblemWithConstraint(
                         lambda x: surrogateModel.bumpiness_measure(
                             x, f_target, LDLt
@@ -871,9 +871,7 @@ def acquire(
             ].T
 
             # Evaluate the surrogate model on the candidate points and sort them
-            fcand[iStart:iEnd], _ = surrogateModel.eval(
-                candidates[iStart:iEnd, :]
-            )
+            fcand[iStart:iEnd], _ = surrogateModel(candidates[iStart:iEnd, :])
             ids = np.argsort(fcand[0:iEnd])
             remevals -= iEnd - iStart
 
@@ -899,26 +897,18 @@ def acquire(
                 def func_continuous_search(x):
                     x_ = xi.copy()
                     x_[cindex] = x
-                    return surrogateModel.eval(x_)[0]
+                    return surrogateModel(x_)[0]
 
                 def dfunc_continuous_search(x):
                     x_ = xi.copy()
                     x_[cindex] = x
                     return surrogateModel.jac(x_)[cindex]
 
-                # def hessp_continuous_search(x, p):
-                #     x_ = xi.copy()
-                #     x_[cindex] = x
-                #     p_ = np.zeros(dim)
-                #     p_[cindex] = p
-                #     return surrogateModel.hessp(x_, p_)[cindex]
-
                 res = minimize(
                     func_continuous_search,
                     xi[cindex],
                     method="L-BFGS-B",
                     jac=dfunc_continuous_search,
-                    # hessp=hessp_continuous_search,
                     bounds=cbounds,
                     options={
                         "maxfun": remevals,
@@ -1025,7 +1015,7 @@ def pareto_front_target(self, paretoFront: np.ndarray) -> np.ndarray:
         assert objdim > 1
 
         # Create a surrogate model for the Pareto front in the objective space
-        paretoModel = RbfModel(RbfType.LINEAR)
+        paretoModel = RbfModel(kernel=RbfKernel.LINEAR)
         k = np.random.choice(objdim)
         paretoModel.update_samples(
             np.array([paretoFront[:, i] for i in range(objdim) if i != k]).T
@@ -1046,13 +1036,13 @@ def pareto_front_target(self, paretoFront: np.ndarray) -> np.ndarray:
         )
 
         def delta_f(tau):
-            tauk, _ = paretoModel.eval(tau)
+            tauk, _ = paretoModel(tau)
             _tau = np.concatenate((tau[0:k], tauk, tau[k:]))
             return -tree.query(_tau)[0]
 
         # Minimize delta_f
         res = differential_evolution(delta_f, boundsPareto)
-        tauk, _ = paretoModel.eval(res.x)
+        tauk, _ = paretoModel(res.x)
         tau = np.concatenate((res.x[0:k], tauk, res.x[k:]))
 
         return tau
@@ -1175,7 +1165,7 @@ def acquire(
         endpoints = np.empty((objdim, dim))
         for i in range(objdim):
             minimumPointProblem = ProblemNoConstraint(
-                lambda x: surrogateModels[i].eval(x)[0], bounds, iindex
+                lambda x: surrogateModels[i](x)[0], bounds, iindex
             )
             res = pymoo_minimize(
                 minimumPointProblem,
@@ -1421,9 +1411,7 @@ def acquire(
         fnondominatedAndBestCandidates = np.concatenate(
             (
                 paretoFront,
-                np.array(
-                    [s.eval(bestCandidates)[0] for s in surrogateModels]
-                ).T,
+                np.array([s(bestCandidates)[0] for s in surrogateModels]).T,
             ),
             axis=0,
         )
@@ -1508,7 +1496,7 @@ def acquire(
         cheapProblem = ProblemWithConstraint(
             self.fun,
             lambda x: np.transpose(
-                [surrogateModels[i].eval(x)[0] for i in range(gdim)]
+                [surrogateModels[i](x)[0] for i in range(gdim)]
             ),
             bounds,
             iindex,

blackboxopt/optimize.py

@@ -31,7 +31,7 @@
     "Haoyu Jia",
     "Weslley S. Pereira",
 ]
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 __deprecated__ = False
 
 from typing import Callable, Optional, Union
@@ -781,9 +781,9 @@ def multistart_stochastic_response_surface(
         if out_local.fx < out.fx:
             out.x[:] = out_local.x
             out.fx = out_local.fx
-        out.samples[
-            out.nfev : out.nfev + out_local.nfev, :
-        ] = out_local.samples
+        out.samples[out.nfev : out.nfev + out_local.nfev, :] = (
+            out_local.samples
+        )
         out.fsamples[out.nfev : out.nfev + out_local.nfev] = out_local.fsamples
         out.nfev = out.nfev + out_local.nfev
 
@@ -898,7 +898,11 @@ def target_value_optimization(
 
     # max value of f
     if surrogateModel.nsamples() - out.nfev > 0:
-        maxf = np.max(surrogateModel.get_fsamples()[: -out.nfev]).item()
+        maxf = np.max(
+            surrogateModel.get_fsamples()[
+                0 : surrogateModel.nsamples() - out.nfev
+            ]
+        ).item()
     else:
         maxf = -np.Inf
     if out.nfev > 0:
@@ -1681,7 +1685,7 @@ def gosac(
         # Evaluate the surrogate at the best candidates
         sCandidates = np.empty((len(bestCandidates), gdim))
         for i in range(gdim):
-            sCandidates[:, i], _ = surrogateModels[i].eval(bestCandidates)
+            sCandidates[:, i], _ = surrogateModels[i](bestCandidates)
 
         # Find the minimum number of constraint violations
         constraintViolation = [

blackboxopt/problem.py

@@ -20,7 +20,7 @@
 __maintainer__ = "Weslley S. Pereira"
 __email__ = "weslley.dasilvapereira@nrel.gov"
 __credits__ = ["Weslley S. Pereira"]
-__version__ = "0.4.1"
+__version__ = "0.4.2"
 __deprecated__ = False
 
 import numpy as np
@@ -161,7 +161,7 @@ def _evaluate(self, X, out):
         out["F"] = np.empty((x.shape[0], self.n_obj))
         for i in range(self.n_obj):
             out["F"][:, i] = np.absolute(
-                self.surrogateModels[i].eval(x)[0] - self.tau[i]
+                self.surrogateModels[i](x)[0] - self.tau[i]
             )
 
 
@@ -185,4 +185,4 @@ def _evaluate(self, X, out):
         x = _dict_to_array(X)
         out["F"] = np.empty((x.shape[0], self.n_obj))
         for i in range(self.n_obj):
-            out["F"][:, i] = self.surrogateModels[i].eval(x)[0]
+            out["F"][:, i] = self.surrogateModels[i](x)[0]