I want to specify the dimension of the unconstrained optimization problem, how do I set that? about pycutest HOT 8 OPEN

Hi @pangtai6,

Good questions, in reply:

1. You can only set the parameters and values printed by pycutest.print_available_sif_params(), so for example for BROYDN7D it tells us that

>>> pycutest.print_available_sif_params('BROYDN7D')
Parameters available for problem BROYDN7D:
N/2 = 5 (int, n=10 original value) 
N/2 = 25 (int, n=50) 
N/2 = 50 (int, n=100) 
N/2 = 250 (int, n=500) 
N/2 = 500 (int, n=1000) 
N/2 = 2500 (int, n=5000) [default]
N/2 = 5000 (int, n=10000) 
End of parameters for problem BROYDN7D

which means we can only set N/2 to one of the values listed:

>>> problem = pycutest.import_problem('BROYDN7D', sifParams={'N/2':250})
>>> problem.n
500

if you want other values of N/2 you have to put them into the BROYDN7D.SIF file (this is how CUTEst is designed).

2. Newton's method is not 'globally convergent' meaning that it needs a 'globalisation strategy' to ensure that it converges from any starting point for nonlinear problems. This can be linesearch (or trust-region, etc) but for linesearch you need a backtracking Armijo linesearch at minimum: a fixed step size is not guaranteed to converge from any starting point for general nonlinear problems (see the Line Search chapter of Nocedal & Wright's Numerical Opimization for details).

from pycutest.

Thank you very much for your clarity. Wish you a happy life.

from pycutest.

Excuse me, why can't I find the local minimum by changing the initial point here, using the line search Newton method?

from pycutest.

For your example code, you need to resample the gradient/Hessian at each iteration after you have taken your step (after line 36), so the new step calculation in the next loop iteration uses the gradient/Hessian at the new value of x.

from pycutest.

When I import the numpy and scipy, the program will report an error. The result is as follows:

exercise2.py.zip
The zip folder contains the code.
Could you help me, please？

from pycutest.

I see that you're running IPython in the Python Console, please change this to your standard Python Interpreter that you installed pycutest with:
https://www.jetbrains.com/help/pycharm/using-consoles.html#python-console-settings

from pycutest.

I see that you're running IPython in the Python Console, please change this to your standard Python Interpreter that you installed pycutest with:
https://www.jetbrains.com/help/pycharm/using-consoles.html#python-console-settings

I tried it out, and the setup in pycharm was ok, I found it was acceptable not to use the "scipy" package alone, and the "pycutest" instruction worked fine; Similarly, the commands in the "scipy" package will work fine without calling "pycutest". There might be a conflict between the two commands.

from pycutest.

I found the problem and corrected the program. Thank you very much for your reply.

from pycutest.