-
Notifications
You must be signed in to change notification settings - Fork 0
/
local_analysis.m
188 lines (179 loc) · 7.81 KB
/
local_analysis.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
%% Gamma local analysis - Egreedy
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.2;
trueepsilon=0.1;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
gammas=linspace(0,1,100);
parfor i=1:1000
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',gammas(i),'alpha',truealpha);
score(i) = agent.imitate_run(trace);
end
plot(gammas,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,epsilon %.1f being generated by gamma x',truegamma,trueepsilon))
xlabel('Gammas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\gamma_egreedy.jpg'));
%% Gamma local analysis - Softmax
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.05;
truetemperature=10;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',truetemperature,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
gammas=linspace(0,1,100);
for i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',truetemperature,'gamma',gammas(i),'alpha',truealpha);
score(i) = agent.imitate_run(trace);
end
plot(gammas,exp(score));
title(sprintf('Likelihood of a trace with gamma 0.5,temperature %.1f being generated by gamma x',truetemperature))
xlabel('Gammas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\gamma_softmax.jpg'));
%% Epsilon local analysis
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.2;
trueepsilon=0.1;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
epsilons=linspace(0,1,100);
parfor i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',epsilons(i),'gamma',truegamma,'alpha',truealpha);
score(i) = agent.imitate_run(trace);
end
plot(epsilons,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,epsilon %.1f being generated by epsilon x',truegamma,trueepsilon))
xlabel('Epsilons')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\epsilon.jpg'));
%% Temperature local analysis
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.2;
truetemperature=10;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',truetemperature,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
temperatures=linspace(5,50,100);
parfor i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',temperatures(i),'gamma',truegamma,'alpha',truealpha);
score(i) = agent.imitate_run(trace);
end
plot(temperatures,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,temperature %.1f being generated by temperature x',truegamma,truetemperature))
xlabel('Temperatures')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\temperatures.jpg'));
%% Alpha local analysis - Egreedy
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.2;
trueepsilon=0.1;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
alphas=linspace(0,1,100);
parfor i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',truegamma,'alpha',alphas(i));
score(i) = agent.imitate_run(trace);
end
plot(alphas,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,alpha %.1f being generated by alpha x',truegamma,truealpha))
xlabel('Alphas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\alphas_epsilon.jpg'));
%% Alpha local analysis - Softmax
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.5;
truealpha=0.2;
truetemperature=10;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',truetemperature,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
alphas=linspace(0,1,100);
parfor i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',truetemperature,'gamma',truegamma,'alpha',alphas(i));
score(i) = agent.imitate_run(trace);
end
plot(alphas,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,alpha %.1f being generated by alpha x',truegamma,truealpha))
xlabel('Alphas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\alphas_softmax.jpg'));
%%
clear
[model, sim, env ] = games.get(2,1);
truegammas=[0.5 0.6 0.7];
alphas=[0.05 0.1 0.15 0.2 0.3 0.4];
epsilons=[0.1];
for g=1:length(truegammas)
for a=1:length(alphas)
for e=1:length(epsilons)
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',epsilons(e),'gamma',truegammas(g),'alpha',alphas(a));
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
gammas=linspace(0,1,100);
eps=epsilons(e)
for i=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',eps,'gamma',gammas(i),'alpha',alphas(a));
score(i) = agent.imitate_run(trace);
end
plot(gammas,exp(score));
title(sprintf('Likelihood of a trace with gamma %.1f,epsilon %.1f,alpha%.2f being generated by gamma x',truegammas(g),epsilons(e),alphas(a)))
xlabel('Gammas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\gamma_epsilon0%d_gamma%d_alpha%d.jpg',e,g,a));
plot(gammas,score);
title(sprintf('Log-Likelihood of a trace with gamma %.1f,epsilon %.1f,alpha%.2f being generated by gamma x',truegammas(g),epsilons(e),alphas(a)))
xlabel('Gammas')
ylabel('Pr(trace)')
saveas(gcf,sprintf('img\\gamma_epsilon0%d_gamma%d_alpha%d_log.jpg',e,g,a));
end
end
end
%% Gamma-Alpha local analysis - Egreedy Contour
clear
[model, sim, env ] = games.get(2,1);
truegamma=0.65;
truealpha=0.05;
trueepsilon=0.2;
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',truegamma,'alpha',truealpha);
[ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
gammas=linspace(0,1,100);
alphas=linspace(0,1,100);
parfor i=1:100
for j=1:100
agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','egreedy','epsilon',trueepsilon,'gamma',gammas(i),'alpha',alphas(j));
score(i,j) = agent.imitate_run(trace);
end
end
contourf(gammas,alphas,score)
title(sprintf('Likelihood of a trace with gamma %.1f,epsilon %.1f being generated by gamma x',truegamma,trueepsilon))
xlabel('Alphas')
ylabel('Gammas')
% saveas(gcf,sprintf('img\\gamma_egreedy.jpg'));
% %%
% clear
% [model, sim, env ] = games.get(2,1);
% truegamma=0.5;
% truealpha=0.2;
% temperatures=[5 10 15 20 25 30];
% for e=1:length(temperatures)
% agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',temperatures(e),'gamma',truegamma,'alpha',truealpha);
% [ trace policytrace ] = analyse.sampling.synthetic_traces( agent, sim, 300 );
% gammas=linspace(0,1,100);
% for i=1:100
% agent = agents.discrete.get('sarsa',env.numstates,env.numactions,'policytype','softmax','temperature',temperatures(e),'gamma',gammas(i),'alpha',truealpha);
% score(i) = agent.imitate_run(trace);
% end
% plot(gammas,exp(score));
% title(sprintf('Likelihood of a trace with gamma 0.5,temperature %.1f being generated by gamma x',temperatures(e)))
% xlabel('Gammas')
% ylabel('Pr(trace)')
% saveas(gcf,sprintf('img\\gamma_temperature%d.jpg',e));
% end