tol=1e-14,
verbose=1,
line_search=True,
callback=cb_tos)
trace_nols = np.array([loss(x) for x in cb_tos.trace_x])
all_trace_nols.append(trace_nols)
all_trace_nols_time.append(cb_tos.trace_time)
out_img.append(tos.x)
cb_pdhg = cp.utils.Trace()
x0 = np.zeros(n_features)
pdhg = cp.minimize_primal_dual(f.f_grad,
x0,
G1.prox,
G2.prox,
callback=cb_pdhg,
max_iter=max_iter,
step_size=step_size,
step_size2=(1. / step_size) / 2,
tol=0,
line_search=False)
trace_pdhg = np.array([loss(x) for x in cb_pdhg.trace_x])
all_trace_pdhg.append(trace_pdhg)
all_trace_pdhg_time.append(cb_pdhg.trace_time)
cb_pdhg_nols = cp.utils.Trace()
x0 = np.zeros(n_features)
pdhg_nols = cp.minimize_primal_dual(f.f_grad,
x0,
G1.prox,
G2.prox,
callback=cb_pdhg_nols,
verbose=1,
callback=cb_tos,
line_search=False,
)
trace_nols = [loss(x, beta) for x in cb_tos.trace_x]
all_trace_nols.append(trace_nols)
all_trace_nols_time.append(cb_tos.trace_time)
cb_pdhg = cp.utils.Trace()
x0 = np.zeros(n_features)
cp.minimize_primal_dual(
f.f_grad,
x0,
g_prox,
h_prox,
callback=cb_pdhg,
max_iter=max_iter,
step_size=step_size,
step_size2=(1.0 / step_size) / 2,
tol=0,
)
trace_pdhg = np.array([loss(x, beta) for x in cb_pdhg.trace_x])
all_trace_pdhg.append(trace_pdhg)
all_trace_pdhg_time.append(cb_pdhg.trace_time)
# .. plot the results ..
f, ax = plt.subplots(2, 3, sharey=False)
xlim = [0.02, 0.02, 0.1]
for i, beta in enumerate(all_betas):
ax[0, i].set_title(r"$\lambda=%s$" % beta)
ax[0, i].imshow(out_img[i], interpolation="nearest", cmap=plt.cm.gray)
max_iter=max_iter,
tol=0,
callback=cb_tos,
line_search=False,
)
trace_nols = [loss(x, beta) for x in cb_tos.trace_x]
all_trace_nols.append(trace_nols)
all_trace_nols_time.append(cb_tos.trace_time)
cb_pdhg = cp.utils.Trace()
cp.minimize_primal_dual(
f.f_grad,
np.zeros(n_features),
g_prox,
h_prox,
callback=cb_pdhg,
max_iter=max_iter,
step_size=step_size,
step_size2=(1. / step_size) / 2,
line_search=False,
)
trace_pdhg = np.array([loss(x, beta) for x in cb_pdhg.trace_x])
all_trace_pdhg.append(trace_pdhg)
all_trace_pdhg_time.append(cb_pdhg.trace_time)
# .. plot the results ..
f, ax = plt.subplots(2, 3, sharey=False)
xlim = [0.02, 0.02, 0.1]
for i, beta in enumerate(all_betas):
ax[0, i].set_title(r"$\lambda=%s$" % beta)
ax[0, i].imshow(out_img[i], interpolation="nearest", cmap=plt.cm.gray)