-
Notifications
You must be signed in to change notification settings - Fork 12
/
analyzeParameterExperiment.m
171 lines (151 loc) · 5.02 KB
/
analyzeParameterExperiment.m
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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
doRun = true; % Set to true to run the experiment; leave false to just load saved results
caseName = '';
% Either run the experiment or load saved results
if doRun
expObj = runExperiment('parameter');
else
expObj = ParameterExperiment(caseName);
end
fprintf('Experiment loaded. %i of %i trials completed\n',...
length(expObj.CompletedTrials), expObj.NTrials);
%%
% Compute the distribution across trials of the convergence and iterations
r = expObj.computeResultsDistributionMetricMultiple({'Convergence','NumIterations'});
%%
% Check overall convergence
convergenceRate = mean(r.Convergence(:));
fprintf('Convergence rate: %g\n',convergenceRate)
if any(~r.Convergence(:))
disp('Trials where the algorithm didn''t converge for at least one independent variable:');
disp(find(any(~r.Convergence,2)));
end
%%
% Build array of independent variables. First row is gamma, second is
% delta0
ivStructs = expObj.CompletedTrials(1).treatmentVariables;
ivs = nan(2,length(ivStructs));
for i = 1:length(ivStructs)
ivs(1,i) = ivStructs(i).gamma;
ivs(2,i) = ivStructs(i).delta0;
end
%%
% Build array for each trial and treatment variable. Columns are: [gamma,
% delta0, numIterations]
d = nan(numel(r.NumIterations),3);
k = size(r.NumIterations,1);
for i = 1:length(ivStructs)
d((i-1)*k+1:i*k,1) = ivStructs(i).gamma;
d((i-1)*k+1:i*k,2) = ivStructs(i).delta0;
d((i-1)*k+1:i*k,3) = r.NumIterations(:,i);
end
%%
% Plot scatter number iterations against delta0, gamma
figure;
scatter3(d(:,1),d(:,2),d(:,3));
xlabel('gamma');
ylabel('delta0');
%%
% Plot surfaces for mean and max number iterations against delta0, gamma
% Build meshgrid
[Xq, Yq] = meshgrid(linspace(min(ivs(1,:)),max(ivs(1,:))),...
linspace(min(ivs(2,:)),max(ivs(2,:))));
figure;
surf(Xq,Yq,griddata(ivs(1,:),ivs(2,:),mean(r.NumIterations,1),Xq,Yq),...
'facecolor','r','facealpha',0.5);
hold on
surf(Xq,Yq,griddata(ivs(1,:),ivs(2,:),max(r.NumIterations,[],1),Xq,Yq),...
'facecolor','b','facealpha',0.5);
xlabel('\gamma');
ylabel('\delta_0');
title('Number of iterations to convergence');
legend('Mean','Max');
%%
% Make 2-d plots
gammaVals = unique(ivs(1,:));
deltaVals = unique(ivs(2,:));
numBins = 4;
% w.r.t gamma
meanY = nan(length(gammaVals),numBins);
maxY = nan(size(meanY));
labels = {};
for i = 1:length(gammaVals)
ind1 = ivs(1,:) == gammaVals(i);
for j=1:numBins
lb = deltaVals(1) + (deltaVals(end)-deltaVals(1))/numBins*(j-1);
ub = deltaVals(1) + (deltaVals(end)-deltaVals(1))/numBins*j;
if j < numBins
ind2 = ivs(2,:) >= lb & ivs(2,:) < ub;
labels{j} = sprintf('$%g \\leq \\delta_0 < %g$',lb,ub);
else
ind2 = ivs(2,:) >= lb & ivs(2,:) <= ub;
labels{j} = sprintf('$%g \\leq \\delta_0 \\leq %g$',lb,ub);
end
meanY(i,j) = mean(mean(r.NumIterations(:,ind1 & ind2)));
maxY(i,j) = max(max(r.NumIterations(:,ind1 & ind2)));
end
end
f = figure;
set(f,'Position',get(f,'Position').*[1 1 1.1 0.7]);
setPageSize(f);
subplot(1,2,1);
semilogy(gammaVals,meanY,'LineWidth',1);
legend(labels,'Interpreter','latex','orientation','vertical','Location','best');
set(gca,'ColorOrderIndex',1)
hold on
semilogy(gammaVals,maxY,'--','LineWidth',0.75,'HandleVisibility','off')
%ylabel('No. Iterations')
%title('Mean and Max Iterations');
xlabel('\gamma');
shrinkWhitespace(gca);
% w.r.t delta_0
meanY = nan(length(deltaVals),numBins);
maxY = nan(size(meanY));
labels = {};
for i = 1:length(deltaVals)
ind1 = ivs(2,:) == deltaVals(i);
for j=1:numBins
lb = gammaVals(1) + (gammaVals(end)-gammaVals(1))/numBins*(j-1);
ub = gammaVals(1) + (gammaVals(end)-gammaVals(1))/numBins*j;
if j < numBins
ind2 = ivs(1,:) >= lb & ivs(1,:) < ub;
labels{j} = sprintf('$%g \\leq \\gamma < %g$',lb,ub);
else
ind2 = ivs(1,:) >= lb & ivs(1,:) <= ub;
labels{j} = sprintf('$%g \\leq \\gamma \\leq %g$',lb,ub);
end
meanY(i,j) = mean(mean(r.NumIterations(:,ind1 & ind2)));
maxY(i,j) = max(max(r.NumIterations(:,ind1 & ind2)));
end
end
subplot(1,2,2);
semilogy(deltaVals,meanY,'LineWidth',1);
legend(labels,'Interpreter','latex','orientation','vertical','Location','best');
set(gca,'ColorOrderIndex',1)
hold on
semilogy(deltaVals,maxY,'--','LineWidth',0.75,'HandleVisibility','off')
%ylabel('No. Iterations')
xlabel('\delta_0 (kWh)');
shrinkWhitespace(gca);
% Save in eps and pdf formats
%print -dpdf -painters figures/parameter
%saveas(f,'figures/parameter','epsc');
%saveas(f,'figures/parameter','png');
%% Helper functions
function shrinkWhitespace(ax)
outerpos = ax.OuterPosition;
ti = ax.TightInset;
left = outerpos(1) + ti(1) + -0.04;
bottom = ax.Position(2)+0.05;
ax_width = outerpos(3) - ti(1) - ti(3) + 0.04;
ax_height = ax.Position(4);
ax.Position = [left bottom ax_width ax_height];
end
function setPageSize(f)
u1 = get(f,'Units');
set(f,'Units','inches');
screenposition = get(f,'Position');
set(f,...
'PaperPosition',[0 0 screenposition(3:4)],...
'PaperSize',screenposition(3:4));
set(f,'Units',u1);
end