Update priors and examples

examples/1_basics/6_priors.py

@@ -20,6 +20,7 @@
     ConfigurationSpace,
     NormalFloatHyperparameter,
     UniformIntegerHyperparameter,
+    UniformFloatHyperparameter,
 )
 from sklearn.datasets import load_digits
 from sklearn.exceptions import ConvergenceWarning
@@ -38,7 +39,7 @@
 
 class MLP:
     @property
-    def configspace(self) -> ConfigurationSpace:
+    def prior_configspace(self) -> ConfigurationSpace:
         # Build Configuration Space which defines all parameters and their ranges.
         # To illustrate different parameter types,
         # we use continuous, integer and categorical parameters.
@@ -100,7 +101,67 @@ def configspace(self) -> ConfigurationSpace:
         )
 
         # Add all hyperparameters at once:
-        cs.add([n_layer, n_neurons, activation, optimizer, batch_size, learning_rate_init])
+        cs.add(
+            [n_layer, n_neurons, activation, optimizer, batch_size, learning_rate_init]
+        )
+
+        return cs
+
+    @property
+    def configspace(self) -> ConfigurationSpace:
+        # Build Configuration Space which defines all parameters and their ranges.
+        # To illustrate different parameter types,
+        # we use continuous, integer and categorical parameters.
+        cs = ConfigurationSpace()
+
+        # We do not have an educated belief on the number of layers beforehand
+        n_layer = UniformIntegerHyperparameter(
+            "n_layer",
+            lower=1,
+            upper=5,
+        )
+
+        # Define network width without a specific prior
+        n_neurons = UniformIntegerHyperparameter(
+            "n_neurons",
+            lower=8,
+            upper=256,
+        )
+
+        # Define activation functions without specific weights
+        activation = CategoricalHyperparameter(
+            "activation",
+            ["logistic", "tanh", "relu"],
+            default_value="relu",
+        )
+
+        # Define optimizer without specific weights
+        optimizer = CategoricalHyperparameter(
+            "optimizer",
+            ["sgd", "adam"],
+            default_value="adam",
+        )
+
+        # Define batch size without specific distribution
+        batch_size = UniformIntegerHyperparameter(
+            "batch_size",
+            16,
+            512,
+            default_value=128,
+        )
+
+        # Define learning rate range without log-normal prior
+        learning_rate_init = UniformFloatHyperparameter(
+            "learning_rate_init",
+            lower=1e-5,
+            upper=1.0,
+            default_value=1e-3,
+        )
+
+        # Add all hyperparameters at once:
+        cs.add(
+            [n_layer, n_neurons, activation, optimizer, batch_size, learning_rate_init]
+        )
 
         return cs
 
@@ -119,8 +180,12 @@ def train(self, config: Configuration, seed: int = 0) -> float:
             )
 
             # Returns the 5-fold cross validation accuracy
-            cv = StratifiedKFold(n_splits=5, random_state=seed, shuffle=True)  # to make CV splits consistent
-            score = cross_val_score(classifier, digits.data, digits.target, cv=cv, error_score="raise")
+            cv = StratifiedKFold(
+                n_splits=5, random_state=seed, shuffle=True
+            )  # to make CV splits consistent
+            score = cross_val_score(
+                classifier, digits.data, digits.target, cv=cv, error_score="raise"
+            )
 
         return 1 - np.mean(score)
 
@@ -140,8 +205,11 @@ def train(self, config: Configuration, seed: int = 0) -> float:
 
     # We define the prior acquisition function, which conduct the optimization using priors over the optimum
     acquisition_function = PriorAcquisitionFunction(
-        acquisition_function=HyperparameterOptimizationFacade.get_acquisition_function(scenario),
+        acquisition_function=HyperparameterOptimizationFacade.get_acquisition_function(
+            scenario
+        ),
         decay_beta=scenario.n_trials / 10,  # Proven solid value
+        prior_configsapce=mlp.prior_configspace,
     )
 
     # We only want one config call (use only one seed in this example)

smac/acquisition/function/prior_acquisition_function.py

@@ -17,6 +17,7 @@
 from smac.model.abstract_model import AbstractModel
 from smac.model.random_forest.abstract_random_forest import AbstractRandomForest
 from smac.utils.logging import get_logger
+from ConfigSpace import ConfigurationSpace
 
 __copyright__ = "Copyright 2022, automl.org"
 __license__ = "3-clause BSD"
@@ -49,6 +50,7 @@ def __init__(
         self,
         acquisition_function: AbstractAcquisitionFunction,
         decay_beta: float,
+        prior_configspace: ConfigurationSpace,
         prior_floor: float = 1e-12,
         discretize: bool = False,
         discrete_bins_factor: float = 10.0,
@@ -58,8 +60,9 @@ def __init__(
         self._functions: list[AbstractAcquisitionFunction] = []
         self._eta: float | None = None
 
-        self._hyperparameters: dict[Any, Configuration] | None = None
         self._decay_beta = decay_beta
+        self._prior_configspace = prior_configspace
+        self._hyperparameters: dict[Any, Configuration] | None = dict(self._prior_configspace)
         self._prior_floor = prior_floor
         self._discretize = discretize
         self._discrete_bins_factor = discrete_bins_factor
@@ -72,10 +75,12 @@ def __init__(
             acquisition_type = self._acquisition_function
 
         self._rescale = isinstance(acquisition_type, (LCB, TS))
-        
+
         # Variables needed to adapt the weighting of the prior
-        self._initial_design_size = None # The amount of datapoints in the initial design
-        self._iteration_number = 1 # The amount of configurations the prior was used in the selection of configurations. It starts at 1
+        self._initial_design_size = (
+            None  # The amount of datapoints in the initial design
+        )
+        self._iteration_number = 1  # The amount of configurations the prior was used in the selection of configurations. It starts at 1
 
     @property
     def name(self) -> str:  # noqa: D102
@@ -103,8 +108,6 @@ def model(self) -> AbstractModel | None:  # noqa: D102
     @model.setter
     def model(self, model: AbstractModel) -> None:
         self._model = model
-        # TODO replace deprecated method
-        self._hyperparameters = model._configspace.get_hyperparameters_dict()
 
         if isinstance(model, AbstractRandomForest):
             if not self._discretize: