-
Notifications
You must be signed in to change notification settings - Fork 1
/
sim_trajectories.m
214 lines (169 loc) · 6.96 KB
/
sim_trajectories.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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
function traj = sim_trajectories(mdlstruct,optstruct,dpstruct,t_max,n_traj,start_x,is_display)
% SIM_TRAJECTORIES Simulate controled trajectories according EDP results
% SIM_TRAJECTORIES(MDLSTRUCT,OPTSTRUCT,DPSTRUCT,T_MAX,N_TRAJ,START_X,IS_DISPLAY)
% simulates N_TRAJ starting from START_X, until time horizon T_MAX. At
% each step, the optimal control is computed thanks to the EDP policy.
% START_X may be a string: 'mean' for the mean point in the dataset,
% 'last' for the last point in the dataset, or a vector.
%
% IS_DISPLAY indicates if a figure containing the mean trajectory is produced.
n_lags = mdlstruct.n_lags;
ind_available_var = mdlstruct.ind_available_var;
ind_current_var = mdlstruct.ind_current_var;
n_dim=mdlstruct.n_dim;
if nargin <7
is_display = 0;
end
n_dim = mdlstruct.n_dim;
n_real_dim=size(mdlstruct.data,2);
x_init = [];
if ischar(start_x)
switch start_x
case 'last'
if max(n_lags)==1
x_init = mdlstruct.data(end,:);
else
for i= 1:size(mdlstruct.data,2)
x_init = [x_init mdlstruct.data(end:-1:end-n_lags(i)+1,:)'];
end
end
case 'mean'
if max(n_lags)==1
x_init = mean(mdlstruct.data);
else
for i= 1:size(mdlstruct.data,2)
x_init = [x_init mean(mdlstruct.data).*ones(1,n_lags(i))];
end
end
end
else
x_init = start_x;
end
X{1} = repmat(x_init, n_traj,1);
tot_value = 0;
tot_unweighted_value=zeros(1,n_dim);
for t=2:t_max
t
[opt_control_tmp,unweighted_opt_value,weighted_opt_value]=td_policy(X{t-1},optstruct,dpstruct,mdlstruct);
if t==2
traj.th_value = weighted_opt_value;
traj.th_unweighted_value = unweighted_opt_value;
end
opt_control{t-1} = opt_control_tmp;
if isfield(mdlstruct,'model')
% in this case the control must be set up to 0 for each variables
% unused in the td learning
tmp_control = zeros(n_traj,n_real_dim);
tmp_control(:,ind_available_var) = opt_control{t-1};
% X{t} = mdlstruct.model(X{t-1},tmp_control,0);
tmp_X = mdlstruct.model(X{t-1},tmp_control,0);
next_X = [];
for i= 1:size(mdlstruct.data,2)
% if n_lags(i)>1
% next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_current_var(i):ind_current_var(i)+n_lags(i)-2)];
% else
% next_X = [next_X tmp_X(:,i)];
% end
if n_lags(i)==1
next_X = [next_X mu(:,i) tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i)))];
else
if n_lags(i)>1
next_X = [next_X mu(:,i) tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i))) X{t-1}(:,ind_available_var(i)+1:ind_available_var(i)+n_lags(i)-2)];
end
end
end
X{t} = next_X;
else
if isfield(mdlstruct,'gp_model')
% X{t} =mdlstruct.gp_model(X{t-1},opt_control{t-1},0);
tmp_X =mdlstruct.gp_model(X{t-1},opt_control{t-1},0);
tmp_X(tmp_X<0) = 0;
next_X = [];
tmp_control = opt_control{t-1};
for i= 1:size(mdlstruct.data,2)
% if n_lags(i)>1
% next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_current_var(i):ind_current_var(i)+n_lags(i)-2)];
%
% else
% next_X = [next_X tmp_X(:,i)];
% end
if n_lags(i)==1
next_X = [next_X tmp_X(:,i)];
% next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i)))];
else
if n_lags(i)== 2
next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i)))];
else
% nlag>2
% next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i)))];
next_X = [next_X tmp_X(:,i) X{t-1}(:,ind_available_var(i)).*(1-tmp_control(:,ind_available_var(i))) X{t-1}(:,ind_available_var(i)+1:ind_available_var(i)+n_lags(i)-2)];
end
end
end
X{t} = next_X;
end
end
[instant_value_tmp{t-1} unweighted_instant_value_tmp{t-1}] = optstruct.reward(X{t-1}(:,ind_available_var),opt_control{t-1},optstruct.weights);
tot_unweighted_value = tot_unweighted_value+unweighted_instant_value_tmp{t-1}*optstruct.discount_factor^(t-2);
tot_value = tot_value+instant_value_tmp{t-1}*optstruct.discount_factor^(t-2);
instant_cum_value(t,1:n_traj) = tot_value;
end
for t=1:t_max-1
data(t,1:n_traj,1:n_real_dim) = X{t}(:,ind_current_var);
control(t,1:n_traj,1:n_dim) =opt_control{t};
end
traj.data = data;
traj.control = control;
traj.value = tot_value;
traj.instant_cum_value = instant_cum_value;
traj.unweighted_value = tot_unweighted_value;
traj.weights =optstruct.weights;
name = mdlstruct.name;
%% display
if is_display
mean_x = squeeze(mean(traj.data,2));
mean_catch = squeeze(mean(traj.data(:,:,ind_available_var).*traj.control,2));
if n_traj>1
mean_v = squeeze(mean(traj.instant_cum_value'));
else
mean_v = squeeze(traj.instant_cum_value');
end
figure
title('Mean trajectory')
hold on
[pplot,nplot]=numSubplots(n_real_dim+1);
for i=1:n_real_dim
subplot(pplot(1),pplot(2),i);
hold on
plot(mean_x(:,i),'b','Linewidth',2);
switch mdlstruct.control_type
case 'rate'
if n_lags(i) ~= 0
plot(mean_catch(:,i),'c','Linewidth',2);
leg_control = [name{i} ' catch'];
end
case 'single'
% to do
case 'global'
% to do
end
legend(name{i},leg_control);
xlabel('Time')
ylabel(name{i})
grid on; box on
end
subplot(pplot(1),pplot(2),n_real_dim+1);
hold on
plot(mean_v,'g','Linewidth',2);
% legend('cumulative reward');
xlabel('Time')
ylabel('Reward')
grid on; box on
end
disp('**********************************************************')
disp(['Theorical value: ' num2str(mean(traj.th_value))])
disp(' ')
disp(['Simulated mean value: ' num2str(mean(traj.value))])
disp(' ')
disp('**********************************************************')
end