def __call__(self, vec_x0, conf=None, fun_smooth=None, fun_smooth_grad=None, fun_a=None, fun_a_grad=None, fun_b=None, fun_b_grad=None, lin_solver=None, status=None): conf = get_default(conf, self.conf) fun_smooth = get_default(fun_smooth, self.fun_smooth) fun_smooth_grad = get_default(fun_smooth_grad, self.fun_smooth_grad) fun_a = get_default(fun_a, self.fun_a) fun_a_grad = get_default(fun_a_grad, self.fun_a_grad) fun_b = get_default(fun_b, self.fun_b) fun_b_grad = get_default(fun_b_grad, self.fun_b_grad) lin_solver = get_default(lin_solver, self.lin_solver) status = get_default(status, self.status) time_stats = {} vec_x = vec_x0.copy() vec_x_last = vec_x0.copy() vec_dx = None if self.log is not None: self.log.plot_vlines(color='r', linewidth=1.0) err0 = -1.0 err_last = -1.0 it = 0 step_mode = 'regular' r_last = None reuse_matrix = False while 1: ls = 1.0 vec_dx0 = vec_dx; i_ls = 0 while 1: tt = time.clock() try: vec_smooth_r = fun_smooth(vec_x) vec_a_r = fun_a(vec_x) vec_b_r = fun_b(vec_x) except ValueError: vec_smooth_r = vec_semismooth_r = None if (it == 0) or (ls < conf.ls_min): output('giving up!') raise else: ok = False else: if conf.semismooth: # Semi-smooth equation. vec_semismooth_r = (nm.sqrt(vec_a_r**2.0 + vec_b_r**2.0) - (vec_a_r + vec_b_r)) else: # Non-smooth equation (brute force). vec_semismooth_r = nm.where(vec_a_r < vec_b_r, vec_a_r, vec_b_r) r_last = (vec_smooth_r, vec_a_r, vec_b_r, vec_semismooth_r) ok = True time_stats['rezidual'] = time.clock() - tt if ok: vec_r = nm.r_[vec_smooth_r, vec_semismooth_r] try: err = nla.norm(vec_r) except: output('infs or nans in the residual:', vec_semismooth_r) output(nm.isfinite(vec_semismooth_r).all()) debug() if self.log is not None: self.log(err, it) if it == 0: err0 = err; break if err < (err_last * conf.ls_on): step_mode = 'regular' break else: output('%s step line search' % step_mode) red = conf.ls_red[step_mode]; output('iter %d, (%.5e < %.5e) (new ls: %e)'\ % (it, err, err_last * conf.ls_on, red * ls)) else: # Failed to compute rezidual. red = conf.ls_red_warp; output('rezidual computation failed for iter %d' ' (new ls: %e)!' % (it, red * ls)) if ls < conf.ls_min: if step_mode == 'regular': output('restore previous state') vec_x = vec_x_last.copy() (vec_smooth_r, vec_a_r, vec_b_r, vec_semismooth_r) = r_last err = err_last reuse_matrix = True step_mode = 'steepest_descent' else: output('linesearch failed, continuing anyway') break ls *= red; vec_dx = ls * vec_dx0; vec_x = vec_x_last.copy() - vec_dx i_ls += 1 # End residual loop. output('%s step' % step_mode) if self.log is not None: self.log.plot_vlines([1], color=self._colors[step_mode], linewidth=0.5) err_last = err; vec_x_last = vec_x.copy() condition = conv_test(conf, it, err, err0) if condition >= 0: break tt = time.clock() if not reuse_matrix: mtx_jac = self.compute_jacobian(vec_x, fun_smooth_grad, fun_a_grad, fun_b_grad, vec_smooth_r, vec_a_r, vec_b_r) else: reuse_matrix = False time_stats['matrix'] = time.clock() - tt tt = time.clock() if step_mode == 'regular': vec_dx = lin_solver(vec_r, mtx=mtx_jac) vec_e = mtx_jac * vec_dx - vec_r lerr = nla.norm(vec_e) if lerr > (conf.eps_a * conf.lin_red): output('linear system not solved! (err = %e)' % lerr) output('switching to steepest descent step') step_mode = 'steepest_descent' vec_dx = mtx_jac.T * vec_r else: vec_dx = mtx_jac.T * vec_r time_stats['solve'] = time.clock() - tt for kv in six.iteritems(time_stats): output('%10s: %7.2f [s]' % kv) vec_x -= vec_dx it += 1 if status is not None: status['time_stats'] = time_stats status['err0'] = err0 status['err'] = err status['condition'] = condition if conf.log.plot is not None: if self.log is not None: self.log(save_figure=conf.log.plot) return vec_x
def __call__(self, vec_x0, conf=None, fun_smooth=None, fun_smooth_grad=None, fun_a=None, fun_a_grad=None, fun_b=None, fun_b_grad=None, lin_solver=None, status=None): conf = get_default(conf, self.conf) fun_smooth = get_default(fun_smooth, self.fun_smooth) fun_smooth_grad = get_default(fun_smooth_grad, self.fun_smooth_grad) fun_a = get_default(fun_a, self.fun_a) fun_a_grad = get_default(fun_a_grad, self.fun_a_grad) fun_b = get_default(fun_b, self.fun_b) fun_b_grad = get_default(fun_b_grad, self.fun_b_grad) lin_solver = get_default(lin_solver, self.lin_solver) status = get_default(status, self.status) time_stats = {} vec_x = vec_x0.copy() vec_x_last = vec_x0.copy() vec_dx = None if self.log is not None: self.log.plot_vlines(color='r', linewidth=1.0) err0 = -1.0 err_last = -1.0 it = 0 step_mode = 'regular' r_last = None reuse_matrix = False while 1: ls = 1.0 vec_dx0 = vec_dx; i_ls = 0 while 1: tt = time.clock() try: vec_smooth_r = fun_smooth(vec_x) vec_a_r = fun_a(vec_x) vec_b_r = fun_b(vec_x) except ValueError: vec_smooth_r = vec_semismooth_r = None if (it == 0) or (ls < conf.ls_min): output('giving up!') raise else: ok = False else: if conf.semismooth: # Semi-smooth equation. vec_semismooth_r = (nm.sqrt(vec_a_r**2.0 + vec_b_r**2.0) - (vec_a_r + vec_b_r)) else: # Non-smooth equation (brute force). vec_semismooth_r = nm.where(vec_a_r < vec_b_r, vec_a_r, vec_b_r) r_last = (vec_smooth_r, vec_a_r, vec_b_r, vec_semismooth_r) ok = True time_stats['residual'] = time.clock() - tt if ok: vec_r = nm.r_[vec_smooth_r, vec_semismooth_r] try: err = nla.norm(vec_r) except: output('infs or nans in the residual:', vec_semismooth_r) output(nm.isfinite(vec_semismooth_r).all()) debug() if self.log is not None: self.log(err, it) if it == 0: err0 = err; break if err < (err_last * conf.ls_on): step_mode = 'regular' break else: output('%s step line search' % step_mode) red = conf.ls_red[step_mode]; output('iter %d, (%.5e < %.5e) (new ls: %e)'\ % (it, err, err_last * conf.ls_on, red * ls)) else: # Failed to compute residual. red = conf.ls_red_warp; output('residual computation failed for iter %d' ' (new ls: %e)!' % (it, red * ls)) if ls < conf.ls_min: if step_mode == 'regular': output('restore previous state') vec_x = vec_x_last.copy() (vec_smooth_r, vec_a_r, vec_b_r, vec_semismooth_r) = r_last err = err_last reuse_matrix = True step_mode = 'steepest_descent' else: output('linesearch failed, continuing anyway') break ls *= red; vec_dx = ls * vec_dx0; vec_x = vec_x_last.copy() - vec_dx i_ls += 1 # End residual loop. output('%s step' % step_mode) if self.log is not None: self.log.plot_vlines([1], color=self._colors[step_mode], linewidth=0.5) err_last = err; vec_x_last = vec_x.copy() condition = conv_test(conf, it, err, err0) if condition >= 0: break tt = time.clock() if not reuse_matrix: mtx_jac = self.compute_jacobian(vec_x, fun_smooth_grad, fun_a_grad, fun_b_grad, vec_smooth_r, vec_a_r, vec_b_r) else: reuse_matrix = False time_stats['matrix'] = time.clock() - tt tt = time.clock() if step_mode == 'regular': vec_dx = lin_solver(vec_r, mtx=mtx_jac) vec_e = mtx_jac * vec_dx - vec_r lerr = nla.norm(vec_e) if lerr > (conf.eps_a * conf.lin_red): output('linear system not solved! (err = %e)' % lerr) output('switching to steepest descent step') step_mode = 'steepest_descent' vec_dx = mtx_jac.T * vec_r else: vec_dx = mtx_jac.T * vec_r time_stats['solve'] = time.clock() - tt for kv in six.iteritems(time_stats): output('%10s: %7.2f [s]' % kv) vec_x -= vec_dx it += 1 if status is not None: status['time_stats'] = time_stats status['err0'] = err0 status['err'] = err status['condition'] = condition if conf.log.plot is not None: if self.log is not None: self.log(save_figure=conf.log.plot) return vec_x