import inspect
from typing import Callable, Any, Optional
[docs]def black_box(
function: Callable[[Any], float],
n_in: int = None,
n_out: int = 1,
fd_method: str = "central",
fd_step: Optional[float] = None,
fd_step_iter: Optional[bool] = None,
) -> Callable[[Any], float]:
"""
Wraps a function as a black box, allowing it to be used in AeroSandbox / CasADi optimization problems.
Obtains gradients via finite differences. Assumes that the function's Jacobian is fully dense, always.
Args:
function:
n_in:
n_out:
fd_method: One of:
- 'forward'
- 'backward'
- 'central'
- 'smoothed'
Returns:
"""
### Grab the signature of the function to be wrapped - we'll need it.
signature = inspect.signature(function)
### Handle default arguments.
if n_in is None:
n_in = len(signature.parameters)
if n_out is None:
n_out = 1
### Add limitations
if n_out > 1:
raise NotImplementedError(
"Black boxes with multiple outputs are not yet supported."
)
### Compute finite-differencing options
fd_options = {}
if fd_step is not None:
fd_options["h"] = fd_step
if fd_step_iter is not None:
fd_options["h_iter"] = fd_step_iter
import casadi as cas
class BlackBox(cas.Callback):
def __init__(
self,
):
cas.Callback.__init__(self)
self.construct(
self.__class__.__name__,
dict(
enable_fd=True,
fd_method=fd_method,
fd_options=fd_options,
),
)
# Number of inputs and outputs
def get_n_in(self):
"""
Number of scalar inputs to the black-box function.
"""
return n_in
def get_n_out(self):
return n_out
# Evaluate numerically
def eval(self, args):
f = function(*args)
if isinstance(f, tuple):
return f
else:
return [f]
# `wrapped_function` is a function with the same call signature as the original function, but with all arguments as positional arguments.
wrapped_function = BlackBox()
def wrapped_function_with_kwargs_support(*args, **kwargs):
"""
This is a function with the same call signature as the original function, allowing both positional and keyword
arguments. Should work identically to the original function:
- Keyword arguments should be optional, and the default values should be the same as the original function.
- Keyword arguments should have the option of being passed as positional arguments, provided they are in the
correct order and all required positional arguments are passed.
- Keyword arguments should be allowed to be passed in any order.
- Positional arguments should be required.
- Positional arguments should have the option of being passed as keyword arguments.
"""
inputs = []
# Check number of positional arguments in the signature
n_positional_args = len(signature.parameters) - len(
signature.parameters.values()
)
n_args = len(signature.parameters)
if len(args) < n_positional_args or len(args) > n_args:
raise TypeError(
f"Takes from {n_positional_args} to {n_args} positional arguments but {len(args)} were given"
)
for i, (name, parameter) in enumerate(signature.parameters.items()):
if i < len(args):
input = args[i]
if name in kwargs:
raise TypeError(
f"Got multiple values for argument '{name}': {input} and {kwargs[name]}"
)
elif name in kwargs:
input = kwargs[name]
else:
if parameter.default is parameter.empty:
raise TypeError(f"Missing required argument '{name}'")
else:
input = parameter.default
# print(input)
inputs.append(input)
return wrapped_function(*inputs)
wrapped_function_with_kwargs_support.wrapped_function = wrapped_function
wrapped_function_with_kwargs_support.wrapper_class = BlackBox
return wrapped_function_with_kwargs_support
if __name__ == "__main__":
### Create a function that's effectively black-box (doesn't use `aerosandbox.numpy`)
[docs] def my_func(
a1,
a2,
k1=4,
k2=5,
k3=6,
):
import math
return math.sin(a1) * math.exp(a2) * math.cos(k1 * k2) + k3
### Now, start an optimization problem
import aerosandbox as asb
import aerosandbox.numpy as np
opti = asb.Opti()
# Wrap our function such that it can be used in an optimization problem.
my_func_wrapped = black_box(function=my_func, fd_method="central")
# Pick some variables to optimize over
m = opti.variable(init_guess=5, lower_bound=3, upper_bound=8)
n = opti.variable(init_guess=5, lower_bound=3, upper_bound=8)
# Minimize the black-box function
opti.minimize(my_func_wrapped(m, a2=3, k2=n))
# Solve
sol = opti.solve()
### Plot the function over its inputs
import matplotlib.pyplot as plt
import aerosandbox.tools.pretty_plots as p
M, N = np.meshgrid(
np.linspace(3, 8, 300),
np.linspace(3, 8, 300),
)
fig, ax = plt.subplots()
p.contour(
M,
N,
np.vectorize(my_func)(M, a2=3, k2=N),
)
p.show_plot()