Adds Transformation class

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## Features
 
+- [#357](https://github.com/pybop-team/PyBOP/pull/357) - Adds `Transformation()` class with `LogTransformation()`, `IdentityTransformation()`, and `ScaledTransformation()`, `ComposedTransformation()` implementations with corresponding examples and tests.
 - [#427](https://github.com/pybop-team/PyBOP/issues/427) - Adds the nbstripout pre-commit hook to remove unnecessary metadata from notebooks.
 - [#327](https://github.com/pybop-team/PyBOP/issues/327) - Adds the `WeightedCost` subclass, defines when to evaluate a problem and adds the `spm_weighted_cost` example script.
 - [#393](https://github.com/pybop-team/PyBOP/pull/383) - Adds Minkowski and SumofPower cost classes, with an example and corresponding tests.

examples/scripts/spm_CMAES.py

@@ -13,12 +13,14 @@
         prior=pybop.Gaussian(6e-06, 0.1e-6),
         bounds=[1e-6, 9e-6],
         true_value=parameter_set["Negative particle radius [m]"],
+        transformation=pybop.LogTransformation(),
     ),
     pybop.Parameter(
         "Positive particle radius [m]",
         prior=pybop.Gaussian(4.5e-06, 0.1e-6),
         bounds=[1e-6, 9e-6],
         true_value=parameter_set["Positive particle radius [m]"],
+        transformation=pybop.LogTransformation(),
     ),
 )
 
@@ -42,7 +44,7 @@
 # Generate problem, cost function, and optimisation class
 problem = pybop.FittingProblem(model, parameters, dataset, signal=signal)
 cost = pybop.SumSquaredError(problem)
-optim = pybop.CMAES(cost, max_iterations=100)
+optim = pybop.CMAES(cost, sigma0=0.25, max_unchanged_iterations=20, max_iterations=50)
 
 # Run the optimisation
 x, final_cost = optim.run()

examples/standalone/cost.py

@@ -21,7 +21,7 @@ class StandaloneCost(pybop.BaseCost):
 
     Methods
     -------
-    __call__(x, grad=None)
+    __call__(x)
         Calculate the cost for a given parameter value.
     """
 
@@ -43,7 +43,7 @@ def __init__(self, problem=None):
         )
         self.x0 = self.parameters.initial_value()
 
-    def _evaluate(self, inputs, grad=None):
+    def _evaluate(self, inputs):
         """
         Calculate the cost for a given parameter value.
 
@@ -54,9 +54,6 @@ def _evaluate(self, inputs, grad=None):
         ----------
         inputs : Dict
             The parameters for which to evaluate the cost.
-        grad : array-like, optional
-            Unused parameter, present for compatibility with gradient-based
-            optimizers.
 
         Returns
         -------

pybop/__init__.py

@@ -55,6 +55,17 @@
 #
 from ._dataset import Dataset
 
+#
+# Transformation classes
+#
+from .transformation.base_transformation import Transformation
+from .transformation.transformations import (
+    IdentityTransformation,
+    ScaledTransformation,
+    LogTransformation,
+    ComposedTransformation,
+)
+
 #
 # Parameter classes
 #

pybop/costs/_likelihoods.py

@@ -39,7 +39,7 @@ def __init__(self, problem: BaseProblem, sigma0: Union[list[float], float]):
         self._offset = -0.5 * self.n_time_data * np.log(2 * np.pi * self.sigma2)
         self._multip = -1 / (2.0 * self.sigma2)
 
-    def _evaluate(self, inputs: Inputs, grad: Union[None, np.ndarray] = None) -> float:
+    def _evaluate(self, inputs: Inputs) -> float:
         """
         Evaluates the Gaussian log-likelihood for the given parameters with known sigma.
         """
@@ -175,7 +175,7 @@ def dsigma_scale(self, new_value):
             raise ValueError("dsigma_scale must be non-negative")
         self._dsigma_scale = new_value
 
-    def _evaluate(self, inputs: Inputs, grad: Union[None, np.ndarray] = None) -> float:
+    def _evaluate(self, inputs: Inputs) -> float:
         """
         Evaluates the Gaussian log-likelihood for the given parameters.
 
@@ -296,17 +296,14 @@ def __init__(self, problem, likelihood, sigma0=None, gradient_step=1e-3):
         ):
             raise ValueError(f"{self.likelihood} must be a subclass of BaseLikelihood")
 
-    def _evaluate(self, inputs: Inputs, grad=None) -> float:
+    def _evaluate(self, inputs: Inputs) -> float:
         """
         Calculate the maximum a posteriori cost for a given set of parameters.
 
         Parameters
         ----------
         inputs : Inputs
             The parameters for which to evaluate the cost.
-        grad : array-like, optional
-            An array to store the gradient of the cost function with respect
-            to the parameters.
 
         Returns
         -------

pybop/costs/_weighted_cost.py

@@ -60,17 +60,14 @@ def __init__(self, *args, weights: Optional[list[float]] = None):
             for cost in self.costs:
                 self.parameters.join(cost.parameters)
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Calculate the weighted cost for a given set of parameters.
 
         Parameters
         ----------
         inputs : Inputs
             The parameters for which to compute the cost.
-        grad : array-like, optional
-            An array to store the gradient of the cost function with respect
-            to the parameters.
 
         Returns
         -------
@@ -90,7 +87,7 @@ def _evaluate(self, inputs: Inputs, grad=None):
                 cost._current_prediction = cost.problem.evaluate(inputs)
             else:
                 cost._current_prediction = self._current_prediction
-            e[i] = cost._evaluate(inputs, grad)
+            e[i] = cost._evaluate(inputs)
 
         return np.dot(e, self.weights)
 

pybop/costs/base_cost.py

@@ -32,6 +32,7 @@ class BaseCost:
 
     def __init__(self, problem: Optional[BaseProblem] = None):
         self.parameters = Parameters()
+        self.transformation = None
         self.problem = problem
         self._fixed_problem = False
         self.set_fail_gradient()
@@ -40,29 +41,27 @@ def __init__(self, problem: Optional[BaseProblem] = None):
             self.parameters.join(self.problem.parameters)
             self.n_outputs = self.problem.n_outputs
             self.signal = self.problem.signal
+            self.transformation = self.parameters.construct_transformation()
             self._fixed_problem = True
 
     @property
     def n_parameters(self):
         return len(self.parameters)
 
-    def __call__(self, inputs: Union[Inputs, list], grad=None):
+    def __call__(self, inputs: Union[Inputs, list]):
         """
         Call the evaluate function for a given set of parameters.
         """
-        return self.evaluate(inputs, grad)
+        return self.evaluate(inputs)
 
-    def evaluate(self, inputs: Union[Inputs, list], grad=None):
+    def evaluate(self, inputs: Union[Inputs, list]):
         """
         Call the evaluate function for a given set of parameters.
 
         Parameters
         ----------
         inputs : Inputs or array-like
             The parameters for which to compute the cost and gradient.
-        grad : array-like, optional
-            An array to store the gradient of the cost function with respect
-            to the parameters.
 
         Returns
         -------
@@ -74,21 +73,23 @@ def evaluate(self, inputs: Union[Inputs, list], grad=None):
         ValueError
             If an error occurs during the calculation of the cost.
         """
-        inputs = self.parameters.verify(inputs)
+        if self.transformation:
+            p = self.transformation.to_model(inputs)
+        inputs = self.parameters.verify(p if self.transformation else inputs)
 
         try:
             if self._fixed_problem:
                 self._current_prediction = self.problem.evaluate(inputs)
 
-            return self._evaluate(inputs, grad)
+            return self._evaluate(inputs)
 
         except NotImplementedError as e:
             raise e
 
         except Exception as e:
             raise ValueError(f"Error in cost calculation: {e}") from e
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Calculate the cost function value for a given set of parameters.
 
@@ -98,9 +99,6 @@ def _evaluate(self, inputs: Inputs, grad=None):
         ----------
         inputs : Inputs
             The parameters for which to evaluate the cost.
-        grad : array-like, optional
-            An array to store the gradient of the cost function with respect
-            to the parameters.
 
         Returns
         -------
@@ -134,7 +132,9 @@ def evaluateS1(self, inputs: Union[Inputs, list]):
         ValueError
             If an error occurs during the calculation of the cost or gradient.
         """
-        inputs = self.parameters.verify(inputs)
+        if self.transformation:
+            p = self.transformation.to_model(inputs)
+        inputs = self.parameters.verify(p if self.transformation else inputs)
 
         try:
             if self._fixed_problem:

pybop/costs/design_costs.py

@@ -65,7 +65,7 @@ def update_simulation_data(self, inputs: Inputs):
         self.problem._target = {key: solution[key] for key in self.problem.signal}
         self.dt = solution["Time [s]"][1] - solution["Time [s]"][0]
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Computes the value of the cost function.
 
@@ -75,8 +75,6 @@ def _evaluate(self, inputs: Inputs, grad=None):
         ----------
         inputs : Inputs
             The parameters for which to compute the cost.
-        grad : array, optional
-            Gradient information, not used in this method.
 
         Raises
         ------
@@ -100,16 +98,14 @@ def __init__(self, problem, update_capacity=False):
         super().__init__(problem, update_capacity)
         self._fixed_problem = False  # keep problem evaluation within _evaluate
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Computes the cost function for the energy density.
 
         Parameters
         ----------
         inputs : Inputs
             The parameters for which to compute the cost.
-        grad : array, optional
-            Gradient information, not used in this method.
 
         Returns
         -------
@@ -157,16 +153,14 @@ def __init__(self, problem, update_capacity=False):
         super().__init__(problem, update_capacity)
         self._fixed_problem = False  # keep problem evaluation within _evaluate
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Computes the cost function for the energy density.
 
         Parameters
         ----------
         inputs : Inputs
             The parameters for which to compute the cost.
-        grad : array, optional
-            Gradient information, not used in this method.
 
         Returns
         -------

pybop/costs/fitting_costs.py

@@ -20,7 +20,7 @@ class RootMeanSquaredError(BaseCost):
     def __init__(self, problem):
         super().__init__(problem)
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Calculate the root mean square error for a given set of parameters.
 
@@ -115,17 +115,14 @@ class SumSquaredError(BaseCost):
     def __init__(self, problem):
         super().__init__(problem)
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Calculate the sum of squared errors for a given set of parameters.
 
         Parameters
         ----------
         inputs : Inputs
             The parameters for which to evaluate the cost.
-        grad : array-like, optional
-            An array to store the gradient of the cost function with respect
-            to the parameters.
 
         Returns
         -------
@@ -418,7 +415,7 @@ def __init__(self, observer: Observer):
         self._observer = observer
         self._fixed_problem = False  # keep problem evaluation within _evaluate
 
-    def _evaluate(self, inputs: Inputs, grad=None):
+    def _evaluate(self, inputs: Inputs):
         """
         Calculate the observer cost for a given set of parameters.
 

pybop/optimisers/base_optimiser.py

@@ -58,13 +58,15 @@ def __init__(
         self.verbose = False
         self.log = dict(x=[], x_best=[], cost=[])
         self.minimising = True
+        self.transformation = None
         self.physical_viability = False
         self.allow_infeasible_solutions = False
         self.default_max_iterations = 1000
         self.result = None
 
         if isinstance(cost, BaseCost):
             self.cost = cost
+            self.transformation = self.cost.transformation
             self.parameters.join(cost.parameters)
             self.set_allow_infeasible_solutions()
             if isinstance(cost, (BaseLikelihood, DesignCost)):
@@ -131,6 +133,9 @@ def set_base_options(self):
         # Set other options
         self.verbose = self.unset_options.pop("verbose", self.verbose)
         self.minimising = self.unset_options.pop("minimising", self.minimising)
+        self.transformation = self.unset_options.pop(
+            "transformation", self.transformation
+        )
         if "allow_infeasible_solutions" in self.unset_options.keys():
             self.set_allow_infeasible_solutions(
                 self.unset_options.pop("allow_infeasible_solutions")

pybop/optimisers/base_pints_optimiser.py

@@ -48,9 +48,6 @@
         self._evaluations = None
         self._iterations = None
 
-        # PyBOP doesn't currently support the PINTS transformation class
-        self._transformation = None
-
         self.pints_optimiser = pints_optimiser
         super().__init__(cost, **optimiser_kwargs)
 
@@ -200,8 +197,8 @@
                 return (L, dl) if self.minimising else (-L, -dl)
         else:
 
-            def f(x, grad=None):
-                return self.cost(x, grad) if self.minimising else -self.cost(x, grad)
+            def f(x):
+                return self.cost(x) if self.minimising else -self.cost(x)
 
         # Create evaluator object
         if self._parallel:
@@ -325,8 +322,8 @@
 
             # Show current parameters
             x_user = self.pints_optimiser.x_guessed()
-            if self._transformation is not None:
-                x_user = self._transformation.to_model(x_user)
+            if self.transformation is not None:
+                x_user = self.transformation.to_model(x_user)
             for p in x_user:
                 print(PintsStrFloat(p))
             print("-" * 40)
@@ -348,8 +345,8 @@
             f = self.pints_optimiser.f_best()
 
         # Inverse transform search parameters
-        if self._transformation is not None:
-            x = self._transformation.to_model(x)
+        if self.transformation is not None:
+            x = self.transformation.to_model(x)
 
         return Result(
             x=x, final_cost=f if self.minimising else -f, n_iterations=self._iterations