Module pytop.statement

Classes

class ProblemStatement

The ProblemStatement class is an abstract class for defining the optimization physics. The objective method must be implemented in the derived class. objective gives the DesignVariables as controls. You can access each design variable by its key like as:

x = design_variables["key"]

The x is a Function object. The objective method must return a AdjFloat of the objective function value.

class Problem(ProblemStatement):
    def objective(self, design_variables):
        x = design_variables["key"]
        # do something with x
        solve(a == L, uh, bc)
        # Compute the objective function with the solution uh for example:
        return assemble(inner(f, uh) * dx)

Expand source code

class ProblemStatement(metaclass=ABCMeta):
    '''The ```ProblemStatement``` class is an abstract class for defining the optimization physics.
    The ```objective``` method must be implemented in the derived class. ```objective``` gives
    the ```DesignVariables``` as controls. You can access each design variable by its key like as:
    ```python
    x = design_variables["key"]
    ```
    The ```x``` is a ```Function``` object. The ```objective``` method must return a ```AdjFloat``` of
    the objective function value.
    ```python
    class Problem(ProblemStatement):
        def objective(self, design_variables):
            x = design_variables["key"]
            # do something with x
            solve(a == L, uh, bc)
            # Compute the objective function with the solution uh for example:
            return assemble(inner(f, uh) * dx)
    ```
    '''

    @abstractmethod
    def objective(self, design_variables: DesignVariables, iter_num: int | None, **kwargs) -> AdjFloat:
        '''The objective function. You must implement this method.

        Args:
            design_variables (DesignVariables): The design variables.
            iter_num (int): The iteration number. Iter_num has None when the optimizer calls the objective in backward mode.
            **kwargs: Optional arguments.

        Returns:
            AdjFloat: The objective function value.
        '''
        raise NotImplementedError()

    def compute_sensitivities(self, 
                              design_variables: DesignVariables, 
                              target: str, 
                              variable_key: str) -> np.ndarray:

        control = Control(design_variables.dict_of_original_functions[variable_key])

        if not hasattr(self, target):
            raise AttributeError(f'The "{target}" is not defined.')
        try :
            compute_gradient(getattr(self, target)(design_variables, None), control)
        except AttributeError:
            # warnings.warn(f'The "{target}" is independent of the variable "{variable_key}".')
            return np.zeros(design_variables[variable_key].vector().size())
        sensitivity = fenics_function_to_np_array(compute_gradient(getattr(self, target)(design_variables, None),
                                       control))
        return sensitivity
    
    def recorder(self, xdmf_file: XDMFFile, variable: Form | Function, function_space: FunctionSpace, name: str, iter_num: int | None):
        '''Record the variable to the xdmf file.

        Args:
            xdmf_file (XDMFFile): The xdmf file.
            variable (Form | Function): The variable to record.
            name (str): The name of the variable.
            iter_num (int): The iteration number.
        '''
        variable = project(variable, function_space)
        variable.rename(name, name)
        if iter_num is not None:
            xdmf_file.write(variable, iter_num)
        pass

Methods

def compute_sensitivities(self, design_variables: DesignVariables, target: str, variable_key: str) ‑> numpy.ndarray

def objective(self, design_variables: DesignVariables, iter_num: int | None, **kwargs) ‑> pyadjoint.adjfloat.AdjFloat

The objective function. You must implement this method.

Args

design_variables : DesignVariables

The design variables.

iter_num : int

The iteration number. Iter_num has None when the optimizer calls the objective in backward mode.

**kwargs

Optional arguments.

Returns

AdjFloat

The objective function value.

def recorder(self, xdmf_file: dolfin.cpp.io.XDMFFile, variable: fenics_adjoint.types.function.Function | dolfin.fem.form.Form, function_space: dolfin.function.functionspace.FunctionSpace, name: str, iter_num: int | None)

Record the variable to the xdmf file.

Args

xdmf_file : XDMFFile

The xdmf file.

variable : Form | Function

The variable to record.

name : str

The name of the variable.

iter_num : int

The iteration number.