Open
Description
Python 3.10, Ipopt version 3.14.16 installed from conda-forge.
Here is a toy convex optimization problem in 24 dimensions
import numpy as np
from cyipopt import minimize_ipopt
np.random.seed(42)
nDim = 24
xBase = np.random.normal(0, 1, nDim)
x0 = np.zeros(nDim)
loss = lambda x: np.linalg.norm(x - xBase)
res = minimize_ipopt(loss, x0, tol=1.0E-5, options={'max_iter': 10000})
The answer it produces is correct, but the error message says that it failed to converge. Is this a bug? I have tried different values of tol and max_iter, but the result is still the same. The max_iter argument works in the sense that it changes the number of iterations the algorithm performs. However, I suspect that minimize_ipopt is ignoring the tol argument, because the accuracy of the result it has found is clearly lower than the tolerance I have requested.
message: b'Maximum number of iterations exceeded (can be specified by an option).'
success: False
status: -1
fun: 2.3146770904170615e-08
x: [ 4.967e-01 -1.383e-01 ... 6.753e-02 -1.425e+00]
nit: 10000
info: status: -1
x: [ 4.967e-01 -1.383e-01 ... 6.753e-02 -1.425e+00]
g: []
obj_val: 2.3146770904170615e-08
mult_g: []
mult_x_L: [ 0.000e+00 0.000e+00 ... 0.000e+00 0.000e+00]
mult_x_U: [ 0.000e+00 0.000e+00 ... 0.000e+00 0.000e+00]
status_msg: b'Maximum number of iterations exceeded (can be specified by an option).'
nfev: 470774
njev: 10002
Metadata
Metadata
Assignees
Labels
No labels