CODEWITHSUNDEEP
pythontuplesminimizationscipy-optimize-minimize
nekovolta
nekovolta

Reputation: 516

Pass additional parameters to an objective function

I am trying to solve a minimization problem following this structure:

res=minimize(objective_fnc, x0, args=param, method='Nelder-Mead)

I need to pass a series of arguments that are not par of the optimization variables but parameters to change the conditions/boundaries of the problem in a more general way. For instance, the variable param is a tuple with the next parameters: param = (p1,p2,p3)

Whenever I call the objective_fnc I need to pass the initial conditions x0 and the tuples with the parameters to use these parameters inside the objective function. For instance, objective_fnc(x0,param)

The problem is that I get this error: objective_fnc() takes 2 positional arguments but 4 were given

I know that if I pass the initial conditions and the parameters (without being part of a tuple, this way: objective_fnc(x0,p1,p2,p3)) it works, but I want to write it in a more simplified way in case I need to pass additional parameters later on.

Here you have minimal reproducible example that works, where all the arguments were passed one by one. But if I change p1,p2,p3 for `param' I get the error.

import numpy as np
from scipy.optimize import minimize

def objective_fnc(x,p1,p2,p3):
    """The Rosenbrock function"""
    return sum(100.0*(x[1:]-x[:-1]**2.0)**2.0 + (1-x[:-1])**2.0)


class CallbackFunctor:
    def __init__(self, obj_fun):
        self.num_calls = 0
        if self.num_calls == 0:
            print('Optimization problem started')
        self.obj_fun = obj_fun
    
    def __call__(self, x):
        fun_val = self.obj_fun(x,p1,p2,p3)
        self.num_calls += 1
        if self.num_calls % 5 == 0:
            print('Work in progress...')
            print(p1)
            print(p2)
            print(p3)
            
cb = CallbackFunctor(objective_fnc)

# Parameters
p1=10
p2=20
p3=30
param=(p1,p2,p3)
x0 = np.array([1.3, 0.7, 0.8, 1.9, 1.2])
res = minimize(objective_fnc, x0, args=(p1,p2,p3), method='nelder-mead', callback=cb, options={'xatol': 1e-8, 'disp': True})

print(res.x)

Upvotes: 1

Views: 955

Answers (2)

jmnguye
jmnguye

Reputation: 437

try to use unpatch with :

objective_fnc(Xo, *bound_limits)

https://stackabuse.com/unpacking-in-python-beyond-parallel-assignment

Upvotes: 0

joni
joni

Reputation: 7157

If your function has the signature objective_func(x, bound_limits), where bounds_limits is a tuple, you need to ensure that unpacking args yields a tuple. Hence, you need to pass a Iterable of a tuple:

bound_limits = (param1,param2,param3,param4,param5,param6,param7,param8)
res = minimize(objective_fnc, x0, args=(bound_limits,), method='Nelder-Mead)

Note that there's no need for the args parameter. You can achieve the same by using a lambda function:

res = minimize(lambda x: objective_fnc(x, bound_limits), x0, method='Nelder-Mead')

Upvotes: 1

Related Questions