Is the convergence the condition for OPTMUM the same max relative gradient as on page 5 of Schoenberg paper (2001)?
1 Answer
0
Here is a verbose version of the convergence criteria for Optmum.
_opgtol = 1e-5; // Model parameters x = { 1.5349650, 0.12200957, 1.9751564, 1.4129555 }; // Function value fmin = -2.1746596; // Gradient g = { 7.1666509e-11, 1.2021525e-09, -2.9703797e-10, 1.0380661e-10 }; // Replace any negative values of 'x' // with their absolute value. If any of these // are less than 1, replace them with 1. x_conv = maxv(abs(x), 1); fmin_conv = maxv(abs(fmin), 1); conv_cmp = (abs(g) .* x_conv) ./ fmin_conv; if conv_cmp < _opgtol; print "we have converged!"; endif;
Your Answer
1 Answer
0
Here is a verbose version of the convergence criteria for Optmum.
_opgtol = 1e-5; // Model parameters x = { 1.5349650, 0.12200957, 1.9751564, 1.4129555 }; // Function value fmin = -2.1746596; // Gradient g = { 7.1666509e-11, 1.2021525e-09, -2.9703797e-10, 1.0380661e-10 }; // Replace any negative values of 'x' // with their absolute value. If any of these // are less than 1, replace them with 1. x_conv = maxv(abs(x), 1); fmin_conv = maxv(abs(fmin), 1); conv_cmp = (abs(g) .* x_conv) ./ fmin_conv; if conv_cmp < _opgtol; print "we have converged!"; endif;