Using Scipy.optimize method='SLSQP' returns initial guess

Question

I try to dig more into optimization of functions depending on multiple variables with scipy

I have a function returning prediction from a data mining tool after calling this tool with a batch file.

def query(x):
    import numpy as np
    file_direc_in="path_to_input_file.csv"
    file_direc_out="path_to_output_file.csv"


    with open(file_direc_in, 'w') as f:
        np.savetxt(f, x, delimiter=';', fmt='%.3f',newline='\r\n')
    f.close()
    os.system("Dataset_query.bat")
    #batch file takes the array i wrote to from input_file and estimates a result
    #afterwards the output will be taken from the output file:
    f = open(file_direc_out,'r')
    out = np.array([[float(f.readlines()[0])]])
    f.close()
    return out


from scipy.optimize import minimize
from calc import query
import numpy as np

x0=np.array([[1.5,50,30]])

bnds = ((1, 2), (0.1, 100), (20, 100))

res=minimize(query,x0,method='SLSQP',bounds=bnds, options={'maxiter': 10 , 'disp': True}, callback=True)

when I run the script I see the loop in my console, but it seems that there aren´t really values tested for my variables and I get the initial guess returned:

Optimization terminated successfully.    (Exit mode 0)
            Current function value: [[ 1636.724]]
            Iterations: 1
            Function evaluations: 5
            Gradient evaluations: 1

although I know that for this problem the minimum lies at x_minimum=[1,0.1,100] with a value of out at about out=400 ( i have to decrease the first and second value of the variable and increase the third value to get a lower out)

What am i doing wrong here?

Answer 1

The Solution in my case was changing the step size, because of the unsmoothness of my prediction function query

res=minimize(query,args=(hist,ana),x0=x0,method='SLSQP',/
bounds=bnds, options={'disp': True ,'eps' : 1e0}) 

In my case searching for local minima does not make sense and I am now searching in integer steps for the minima.

according to @ali_m searching for a global minimum basinhopping could be used instead. I will give it a try in the next days