random_search optimiser added in the pints framework (#580)

* random_search optimiser added in the pints framework * style: pre-commit fixes * description texts updated * style: pre-commit fixes * random_search updated * population size input in random search modified * example updated * unit tests added for randomsearch * style: pre-commit fixes * none type boundary handling modified * style: pre-commit fixes * updated * updated * unit tests updated * unit test updated * unit tests modified * updated * style: pre-commit fixes * randomsearch modified * style: pre-commit fixes * boundary logic updated * unit tests updated * unit tests updated * unit test changed * style: pre-commit fixes * fix: RandomSearch with multistart * unit test updated * unit test modified * suggested changes incorporated * suggested changes updated * unit tests for RandomSearch added * style: pre-commit fixes * unit tests modified * style: pre-commit fixes * changelog updated * unit tests added * style: pre-commit fixes * unit test modified * Apply suggestions from code review * Updates unit tests, upper pin to BPX, bugfix _cuckoo and _random_search for population_size setting. --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Brady Planden <[email protected]> Co-authored-by: Brady Planden <[email protected]>
pybop-team · Dec 21, 2024 · a15f7c4 · a15f7c4
1 parent c637f62
commit a15f7c4
Show file tree

Hide file tree

Showing 7 changed files with 311 additions and 7 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -22,6 +22,7 @@
 
 ## Optimisations
 
+- [#580](https://github.com/pybop-team/PyBOP/pull/580) - Random Search optimiser is implimented.
 - [#588](https://github.com/pybop-team/PyBOP/pull/588) - Makes `minimising` a property of `BaseOptimiser` set by the cost class.
 - [#512](https://github.com/pybop-team/PyBOP/pull/513) - Refactors `LogPosterior` with attributes pointing to composed likelihood object.
 - [#551](https://github.com/pybop-team/PyBOP/pull/551) - Refactors Optimiser arguments, `population_size` and `max_iterations` as default args, improves optimiser docstrings

diff --git a/examples/scripts/comparison_examples/random_search.py b/examples/scripts/comparison_examples/random_search.py
@@ -0,0 +1,73 @@
+import numpy as np
+
+import pybop
+
+# Define model
+parameter_set = pybop.ParameterSet.pybamm("Chen2020")
+parameter_set.update(
+    {
+        "Negative electrode active material volume fraction": 0.7,
+        "Positive electrode active material volume fraction": 0.67,
+    }
+)
+model = pybop.lithium_ion.SPM(parameter_set=parameter_set)
+
+# Fitting parameters
+parameters = pybop.Parameters(
+    pybop.Parameter(
+        "Negative electrode active material volume fraction",
+        bounds=[0.4, 0.75],
+        initial_value=0.41,
+    ),
+    pybop.Parameter(
+        "Positive electrode active material volume fraction",
+        bounds=[0.4, 0.75],
+        initial_value=0.41,
+    ),
+)
+experiment = pybop.Experiment(
+    [
+        (
+            "Discharge at 0.5C for 3 minutes (4 second period)",
+            "Charge at 0.5C for 3 minutes (4 second period)",
+        ),
+    ]
+)
+values = model.predict(initial_state={"Initial SoC": 0.7}, experiment=experiment)
+
+sigma = 0.002
+corrupt_values = values["Voltage [V]"].data + np.random.normal(
+    0, sigma, len(values["Voltage [V]"].data)
+)
+
+# Form dataset
+dataset = pybop.Dataset(
+    {
+        "Time [s]": values["Time [s]"].data,
+        "Current function [A]": values["Current [A]"].data,
+        "Voltage [V]": corrupt_values,
+    }
+)
+
+# Generate problem, cost function, and optimisation class
+problem = pybop.FittingProblem(model, parameters, dataset)
+cost = pybop.GaussianLogLikelihood(problem, sigma0=sigma * 4)
+optim = pybop.Optimisation(
+    cost,
+    optimiser=pybop.RandomSearch,
+    max_iterations=100,
+)
+
+results = optim.run()
+
+# Plot the timeseries output
+pybop.plot.quick(problem, problem_inputs=results.x, title="Optimised Comparison")
+
+# Plot convergence
+pybop.plot.convergence(optim)
+
+# Plot the parameter traces
+pybop.plot.parameters(optim)
+
+# Plot the cost landscape with optimisation path
+pybop.plot.contour(optim, steps=10)
diff --git a/pybop/__init__.py b/pybop/__init__.py
@@ -124,6 +124,7 @@
 #
 
 from .optimisers._cuckoo import CuckooSearchImpl
+from .optimisers._random_search import RandomSearchImpl
 from .optimisers._adamw import AdamWImpl
 from .optimisers._gradient_descent import GradientDescentImpl
 from .optimisers.base_optimiser import BaseOptimiser, OptimisationResult, MultiOptimisationResult
@@ -142,6 +143,7 @@
     SNES,
     XNES,
     CuckooSearch,
+    RandomSearch,
     AdamW,
 )
 from .optimisers.optimisation import Optimisation

diff --git a/pybop/optimisers/_cuckoo.py b/pybop/optimisers/_cuckoo.py
@@ -54,7 +54,6 @@ def __init__(self, x0, sigma0=0.05, boundaries=None, pa=0.25):
         self._dim = len(x0)
 
         # Population size and abandon rate
-        self._n = self._population_size
         self._pa = pa
         self.step_size = self._sigma0
         self.beta = 1.5
@@ -68,14 +67,14 @@ def __init__(self, x0, sigma0=0.05, boundaries=None, pa=0.25):
             self._nests = np.random.uniform(
                 low=self._boundaries.lower(),
                 high=self._boundaries.upper(),
-                size=(self._n, self._dim),
+                size=(self._population_size, self._dim),
             )
         else:
             self._nests = np.random.normal(
-                self._x0, self._sigma0, size=(self._n, self._dim)
+                self._x0, self._sigma0, size=(self._population_size, self._dim)
             )
 
-        self._fitness = np.full(self._n, np.inf)
+        self._fitness = np.full(self._population_size, np.inf)
 
         # Initialise best solutions
         self._x_best = np.copy(x0)
@@ -112,7 +111,7 @@ def tell(self, replies):
         self._iterations += 1
 
         # Compare cuckoos with current nests
-        for i in range(self._n):
+        for i in range(self._population_size):
             f_new = replies[i]
             if f_new < self._fitness[i]:
                 self._nests[i] = self.cuckoos[i]
@@ -122,7 +121,7 @@ def tell(self, replies):
                     self._x_best = self.cuckoos[i]
 
         # Abandon some worse nests
-        n_abandon = int(self._pa * self._n)
+        n_abandon = int(self._pa * self._population_size)
         worst_nests = np.argsort(self._fitness)[-n_abandon:]
         for idx in worst_nests:
             self.abandon_nests(idx)

diff --git a/pybop/optimisers/_random_search.py b/pybop/optimisers/_random_search.py
@@ -0,0 +1,111 @@
+import numpy as np
+from pints import PopulationBasedOptimiser
+
+
+class RandomSearchImpl(PopulationBasedOptimiser):
+    """
+    Random Search (RS) optimisation algorithm.
+    This algorithm explores the parameter space by randomly sampling points.
+
+    The algorithm does the following:
+    1. Initialise a population of solutions.
+    2. At each iteration, generate `n` number of random positions within boundaries.
+    3. Evaluate the quality/fitness of the positions.
+    4. Replace the best position with improved position if found.
+
+    Parameters:
+        population_size (optional): Number of solutions to evaluate per iteration.
+
+    References:
+    The Random Search algorithm implemented in this work is based on principles outlined
+    in "Introduction to Stochastic Search and Optimization: Estimation, Simulation, and
+    Control" by Spall, J. C. (2003).
+
+    The implementation inherits from the PINTS PopulationOptimiser.
+    """
+
+    def __init__(self, x0, sigma0=0.05, boundaries=None):
+        super().__init__(x0, sigma0, boundaries=boundaries)
+
+        # Problem dimensionality
+        self._dim = len(x0)
+
+        # Initialise best solution
+        self._x_best = np.copy(x0)
+        self._f_best = np.inf
+        self._running = False
+        self._ready_for_tell = False
+
+    def ask(self):
+        """
+        Returns a list of positions to evaluate in the optimiser-space.
+        """
+        self._ready_for_tell = True
+        self._running = True
+
+        # Generate random solutions
+        if self._boundaries:
+            self._candidates = np.random.uniform(
+                low=self._boundaries.lower(),
+                high=self._boundaries.upper(),
+                size=(self._population_size, self._dim),
+            )
+            return self._candidates
+
+        self._candidates = np.random.normal(
+            self._x0, self._sigma0, size=(self._population_size, self._dim)
+        )
+        return self.clip_candidates(self._candidates)
+
+    def tell(self, replies):
+        """
+        Receives a list of cost function values from points previously specified
+        by `self.ask()`, and updates the optimiser state accordingly.
+        """
+        if not self._ready_for_tell:
+            raise RuntimeError("ask() must be called before tell().")
+
+        # Evaluate solutions and update the best
+        for i in range(self._population_size):
+            f_new = replies[i]
+            if f_new < self._f_best:
+                self._f_best = f_new
+                self._x_best = self._candidates[i]
+
+    def running(self):
+        """
+        Returns ``True`` if the optimisation is in progress.
+        """
+        return self._running
+
+    def x_best(self):
+        """
+        Returns the best parameter values found so far.
+        """
+        return self._x_best
+
+    def f_best(self):
+        """
+        Returns the best score found so far.
+        """
+        return self._f_best
+
+    def name(self):
+        """
+        Returns the name of the optimiser.
+        """
+        return "Random Search"
+
+    def clip_candidates(self, x):
+        """
+        Clip the input array to the boundaries if available.
+        """
+        if self._boundaries:
+            x = np.clip(x, self._boundaries.lower(), self._boundaries.upper())
+        return x
+
+    def _suggested_population_size(self):
+        """
+        Returns a suggested population size based on the dimension of the parameter space.
+        """
+        return 4 + int(3 * np.log(self._n_parameters))
diff --git a/pybop/optimisers/pints_optimisers.py b/pybop/optimisers/pints_optimisers.py
@@ -10,6 +10,7 @@
     BasePintsOptimiser,
     CuckooSearchImpl,
     GradientDescentImpl,
+    RandomSearchImpl,
 )
 
 
@@ -652,3 +653,73 @@ def __init__(
             parallel,
             **optimiser_kwargs,
         )
+
+
+class RandomSearch(BasePintsOptimiser):
+    """
+    Adapter for the Random Search optimiser in PyBOP.
+
+    Random Search is a simple optimisation algorithm that samples parameter sets randomly
+    within the given boundaries and identifies the best solution based on fitness.
+
+    This optimiser has been implemented for benchmarking and comparisons, convergence will be
+    better with one of other optimisers in the majority of cases.
+
+    Parameters
+    ----------
+    cost : callable
+        The cost function to be minimized.
+    max_iterations : int, optional
+        Maximum number of iterations for the optimisation.
+    min_iterations : int, optional (default=2)
+        Minimum number of iterations before termination.
+    max_unchanged_iterations : int, optional (default=15)
+        Maximum number of iterations without improvement before termination.
+    multistart : int, optional (default=1)
+        Number of optimiser restarts from randomly sample position. These positions
+        are sampled from the priors.
+    parallel : bool, optional (default=False)
+        Whether to run the optimisation in parallel.
+    **optimiser_kwargs : optional
+        Valid PINTS option keys and their values, for example:
+        x0 : array_like
+            Initial position from which optimisation will start.
+        population_size : int
+            Number of solutions to evaluate per iteration.
+        bounds : dict
+            A dictionary with 'lower' and 'upper' keys containing arrays for lower and
+            upper bounds on the parameters.
+        absolute_tolerance : float
+            Absolute tolerance for convergence checking.
+        relative_tolerance : float
+            Relative tolerance for convergence checking.
+        max_evaluations : int
+            Maximum number of function evaluations.
+        threshold : float
+            Threshold value for early termination.
+
+    See Also
+    --------
+    pybop.RandomSearchImpl : PyBOP implementation of Random Search algorithm.
+    """
+
+    def __init__(
+        self,
+        cost,
+        max_iterations: int = None,
+        min_iterations: int = 2,
+        max_unchanged_iterations: int = 15,
+        multistart: int = 1,
+        parallel: bool = False,
+        **optimiser_kwargs,
+    ):
+        super().__init__(
+            cost,
+            RandomSearchImpl,
+            max_iterations,
+            min_iterations,
+            max_unchanged_iterations,
+            multistart,
+            parallel,
+            **optimiser_kwargs,
+        )