h="l1",
l_reg=l_reg,
L=L,
display=0)
x2, f2, hist2 = p0b.prox_0bfgs(f,
gf,
x0,
h="l1",
l_reg=l_reg,
L=L,
display=0,
line_search=True)
x3, f3, hist3 = p0s.prox_0sr1(f,
gf,
x0,
h="l1",
l_reg=l_reg,
L=L,
display=0)
x4, f4, hist4 = p0s.prox_0sr1(f,
gf,
x0,
h="l1",
l_reg=l_reg,
L=L,
display=0,
line_search=True)
x5, f5, hist5 = p0s.prox_0sr1(f,
gf,
x0,
h="l1",
gf,
x0,
h="l1",
l_reg=l_reg,
display=0)
x2, f2, hist2 = p0b.prox_0bfgs(f,
gf,
x0,
h="l1",
l_reg=l_reg,
display=0,
line_search=True)
x4, f4, hist4 = p0s.prox_0sr1(f,
gf,
x0,
h="l1",
l_reg=l_reg,
display=0,
line_search=True)
x6, f6, hist6 = p0s.prox_0sr1(f,
gf,
x0,
h="l1",
l_reg=l_reg,
display=0,
exact=True,
line_search=True)
timing += np.array([
hist1.times[-1], hist2.times[-1], 0, hist4.times[-1], 0,
hist6.times[-1]
])
# set up problem 5
x0, x0_lbfgsb, f, f_lbfgsb, gf, gf_lbfgsb, bounds, bounds_lbfgsb, l_reg = setup_problem5(
)
# run Prox0BFGS and Prox0SR1 comparison
x1, f1, hist1 = p0b.prox_0bfgs(f,
gf,
x0,
h="l1_box",
l_reg=l_reg,
bounds=bounds,
display=0,
tol=1e-5)
x3, f3, hist3 = p0s.prox_0sr1(f,
gf,
x0,
h="l1_box",
l_reg=l_reg,
bounds=bounds,
display=0,
tol=1e-5)
# compute optimal solution to high precision
xopt, fopt, hist = p0b.prox_0bfgs(f,
gf,
x0,
h="l1_box",
l_reg=l_reg,
bounds=bounds,
line_search=True,
tol=1e-7,
tol_ssn=1e-12)
# compute ProxGrad and LBFGSB for plotting
x7, f7, hist7 = pg.prox_grad(f,