def bencher_hessian(cost_name, ndim, method_name, x0):
cost_function, cost_function_prime, hessian = mk_costs(ndim)[0][cost_name]
method = methods[method_name]
f_prime = CountingFunction(cost_function_prime)
hessian = CountingFunction(hessian, counter=f_prime.counter)
f = LoggingFunction(cost_function, counter=f_prime.counter)
method(f, x0, jac=f_prime, hess=hessian)
this_costs = np.array(f.all_f_i)
return this_costs, np.array(f.counts)
(mk_quad(.02), bfgs),
(mk_gauss(.02), bfgs),
((rosenbrock, rosenbrock_prime, rosenbrock_hessian),
bfgs),
(mk_quad(.02), powell),
(mk_gauss(.02), powell),
((rosenbrock, rosenbrock_prime, rosenbrock_hessian),
powell),
(mk_gauss(.02), nelder_mead),
((rosenbrock, rosenbrock_prime, rosenbrock_hessian),
nelder_mead),
)):
# Compute a gradient-descent
x_i, y_i = 1.6, 1.1
counting_f_prime = CountingFunction(f_prime)
counting_hessian = CountingFunction(hessian)
logging_f = LoggingFunction(f, counter=counting_f_prime.counter)
all_x_i, all_y_i, all_f_i = optimizer(np.array([x_i, y_i]),
logging_f, counting_f_prime,
hessian=counting_hessian)
# Plot the contour plot
if not max(all_y_i) < y_max:
x_min *= 1.2
x_max *= 1.2
y_min *= 1.2
y_max *= 1.2
x, y = np.mgrid[x_min:x_max:100j, y_min:y_max:100j]
x = x.T
y = y.T