Transform distributions

CHANGELOG.md

@@ -6,6 +6,8 @@
 
 ## Optimisations
 
+- [#925](https://github.com/pybop-team/PyBOP/pull/925) - Add `UnboundedDistribution` and the `get_transformed_distribution` functionality.
+
 ## Bug Fixes
 
 - [#915](https://github.com/pybop-team/PyBOP/pull/915) - Fixes axis labels for non-standard domain names, adds `Dataset` length property and adds `kind` property to `Interpolant`.

examples/notebooks/getting_started/using_transformations.ipynb

@@ -318,7 +318,9 @@
     "problem.parameters[\"Positive electrode active material volume fraction\"] = (\n",
     "    pybop.Parameter(\n",
     "        initial_value=old_param.initial_value,\n",
-    "        bounds=old_param.bounds,\n",
+    "        distribution=pybop.LogUniform(\n",
+    "            lower=old_param.bounds[0], upper=old_param.bounds[1]\n",
+    "        ),\n",
     "        transformation=pybop.LogTransformation(),\n",
     "    )\n",
     ")\n",

examples/scripts/getting_started/maximum_a_posteriori.py

@@ -41,12 +41,10 @@
         "Negative electrode active material volume fraction": pybop.Parameter(
             distribution=pybop.Uniform(0.3, 0.8),
             initial_value=0.653,
-            transformation=pybop.LogTransformation(),
         ),
         "Positive electrode active material volume fraction": pybop.Parameter(
             distribution=pybop.Uniform(0.3, 0.8),
             initial_value=0.657,
-            transformation=pybop.LogTransformation(),
         ),
     }
 )

examples/scripts/getting_started/monte_carlo_sampling.py

@@ -1,3 +1,4 @@
+import numpy as np
 import pybamm
 
 import pybop
@@ -32,11 +33,11 @@
 parameter_values.update(
     {
         "Negative electrode active material volume fraction": pybop.Parameter(
-            distribution=pybop.Gaussian(0.68, 0.02),
+            distribution=pybop.LogNormal(np.log(0.68), 0.02),
             transformation=pybop.LogTransformation(),
         ),
         "Positive electrode active material volume fraction": pybop.Parameter(
-            distribution=pybop.Gaussian(0.65, 0.02),
+            distribution=pybop.LogNormal(np.log(0.65), 0.02),
             transformation=pybop.LogTransformation(),
         ),
     }

pybop/__init__.py

@@ -49,7 +49,7 @@
 # Parameter classes
 #
 from .parameters.parameter import Parameter, Parameters
-from .parameters.distributions import Distribution, Gaussian, Uniform, Exponential, JointDistribution
+from .parameters.distributions import Distribution, Exponential, Gaussian, JointDistribution, LogNormal, LogUniform, Unbounded, Uniform
 from .parameters.multivariate_distributions import MultivariateNonparametric, MultivariateUniform, MultivariateGaussian, MultivariateLogNormal, MarginalDistribution
 
 #

pybop/optimisers/base_optimiser.py

@@ -144,8 +144,6 @@ def run(self) -> "OptimisationResult":
         results = []
         for i in range(self._multistart):
             if i >= 1:
-                if self.problem.parameters.distribution is None:
-                    raise RuntimeError("Distributions must be provided for multi-start")
                 initial_values = self.problem.parameters.sample_from_distribution(1)[0]
                 self.problem.parameters.update(initial_values=initial_values)
                 self._set_up_optimiser()