optim_experiments/acc_nesterov.m at master · windows7lover/optim_experiments · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
% comparison with nesterov

%% Add path automatically
addpath(genpath('../algorithms/'))
addpath(genpath('../classif'))
clear all;
clc;
% close all;
window_size = 10;

backtracking = true;
% backtracking = false;

%% Dataset used in the paper

% dataset = 'sonar'; approxL = false; nIte = 50;
% dataset = 'madelon'; approxL = false; nIte = 100;
dataset = 'sido0'; approxL = false; nIte = 80;

% load dataset
precent_test = 0;
[ paramFunction.X,paramFunction.y,nFeatures,nPoints,paramFunction.Xtest,paramFunction.ytest,nPointsTest] = load_data(dataset, precent_test);

paramFunction.lambda = 0;
paramFunction.approxL = approxL;
load_function

reg = 1e-3*finfo.L; reg_name = 'well_cond';  % well conditionned
% reg = 1e-6*finfo.L; reg_name = 'regular_cond'; % normal conditionned
% reg = 1e-9*finfo.L; reg_name = 'bad_cond'; % badly conditionned


% reg = 1e-2*finfo.L; reg_name = 'comp_nesterov_m4';  % well conditionned
paramFunction.lambda = reg;

load_function;

solve_problem = true;
% solve_problem = false;

%% Algo parameters
nIterTol = NaN;


%%

finfo.mu = 10*finfo.mu ;

if(solve_problem)
    display('Solving with BFGS')
    algoparam.minFuncOpt.Method = 'lbfgs';
%     if(~determ)
%         finfo.f = problem.cost;
%         finfo.fp = problem.full_grad;
%     end
    [xstar,~,~,~] = dokiter_minfunc(algoparam,[],1000,finfo);
    finfo.xstar = xstar;
    fstar = finfo.f(xstar)-(1/(2*reg))*norm(finfo.fp(xstar))^2;
    finfo.fstar = fstar;
    problem.xstar = xstar;
    problem.fstar = fstar;
end


%%

algo_determ;
algoCell_vanilla = {nest,nest_sc};
algoCell_online = {gradient,nest,nest_sc};
% algoCell = {nest};

nAlgo = length(algoCell);
legendCell = cell(1,nAlgo);
lambda = 1e-8; lambda_name = 'lambda_1e-8';

for i=1:length(algoCell_vanilla)
    algoCell_vanilla{i}.name = ['\textbf{(V)} ', algoCell_vanilla{i}.name];
    algoCell_vanilla{i}.param.backtracking = backtracking;
    algoCell_vanilla{i}.linestyle = '--';
end

for i=1:length(algoCell_online)
    algo = algoCell_online{i};
    param = algo.param;

    param.backtracking = backtracking;
    param.dorna = true;
    param.online = true;
    param.window_size = window_size;
    param.lambda = lambda;

    algoCell_online{i}.param = param;
    algoCell_online{i}.name = ['\textbf{(O)} ', algoCell_online{i}.name];
    algoCell_online{i}.linestyle = '-';
end


algoCell_all = {algoCell_vanilla{:}, algoCell_online{:}};
warning off
paramalgo = {}
for i=1:length(algoCell_all)
    i
    algo = algoCell_all{i};
    param = algo.param;
    tic;
    [~,fval,~,paramalgo{i}] = algo.algo(param,nIte,finfo,nIterTol);
    time = toc;
    algoCell_all{i}.fval = fval;
    algoCell_all{i}.time = time;
end
warning on

algoCell_all{1}.name = 'Nest (convex)';
algoCell_all{2}.name = 'Nest (str convex)';
algoCell_all{3}.name = 'Online RNA';
algoCell_all{4}.name = 'ORNA+Nest (convex)';
algoCell_all{5}.name = 'ORNA+Nest (str convex)';
% algoCell_all{1}.name = ''
% algoCell_all{1}.name = ''

for i=1:length(algoCell_all)
    legendCell{i} = algoCell_all{i}.name;
end

%% plots
fs = 16;
lw = 2;

figure
for i=1:length(algoCell_all)
    algo = algoCell_all{i};
    iter = (1:length(algo.fval))-1;
    plot_algo(algo,lw,{});
    hold on
end
legend(legendCell,'fontsize',fs,'box','off','location','sw');
set(gca,'fontsize',fs)
axis([0 80 1e-16, 1])
xlabel('Iteration (Gradient oracle calls)','interpreter','latex')
ylabel('$f(x)-f(x^*)$','interpreter','latex')
set(gca,'fontsize',fs,'color','none')