Callback is not called with local minima of initial guess in basinhopping algorithm

My issue is about …

As far as I understand, callback function is called when a local minima is found. When I use basinhopping global minimization algorithm, I have observed that the callback function is not called when local minima is found for initial guess. I have tried to reproduce this issue with the following example:

from scipy.optimize import basinhopping
import numpy as np

def func(x):
    print("Objective function count: " + str(func.counter))
    func.counter += 1
    return np.random.rand(1)*1000

def callback_func(x, f, accepted):
    print("Callback count: " + str(callback_func.counter))
    callback_func.counter += 1

func.counter = 0
callback_func.counter = 0    
x0 = 2000

minimizer_kwargs = {"method": "COBYLA", "options":{"maxiter": 1}}
optimizer = basinhopping(func, x0, minimizer_kwargs=minimizer_kwargs, niter=2, callback=callback_func)

It produces following result:

Objective function count: 0
Objective function count: 1
Callback count: 0
Objective function count: 2
Callback count: 1

For the above mentioned example, both callback function and the objective function should have been called thrice but callback is called only twice.

Scipy/Numpy/Python version information:

1.2.1 1.16.4 sys.version_info(major=3, minor=7, micro=4, releaselevel='final', serial=0)

[andyfaff]

@rgommers, @prince2110, I think I’ve got to the bottom of this.

• the callback function is indeed called once per basinhopping iteration.
• However, when the BasinHoppingRunner is constructed, there is an initial minimisation. This initial minimisation is not set to the callback function. As the documentation is written it probably should be, but that initial minimisation would not be after a basinhopping step had been taken. At the worst this would result in niter+1 callbacks for niter steps.

Sorry for closing the issue prematurely, @prince2110 is correct here.

[andyfaff]

I’ve already done one, #13029.