-
Notifications
You must be signed in to change notification settings - Fork 33
/
dirichlet_process.m
226 lines (161 loc) · 6.09 KB
/
dirichlet_process.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
215
216
217
218
219
220
221
222
223
224
225
226
%--------------------------------------------------
% visualization of stick-breaking process
%
%
% Written by Richard Xu: [email protected]
% July 2014
%--------------------------------------------------
function dirichlet_process()
% alpah : DP's concentration factor
% K : number of sticks (Large K}
% N_Iter : number of times G ~ DP(alpha, H) is drawn
% init_parts : some initial partiion (just so partitions not too small
% x_range : x range (for plotting)
% max_p : max_p (maximum probability, for plotting)
clear all;
clc;
alpha = 3;
K = 50;
N_Iter = 1000;
CASE = 1;
if CASE == 1
%-----------------------------------------------
% H is gaussian
%-----------------------------------------------
init_parts = -3:2:3;
x_range = [-4 4];
max_p = 0.7;
H_pdf = @normpdf;
H_cdf = @normcdf;
H_rand = @nrand_func;
else
%-----------------------------------------------
% H is gamma
%-----------------------------------------------
init_parts = 0:2:8;
x_range = [0 10];
max_p = 0.7;
H_pdf = @gampdf_func;
H_cdf = @gamcdf_func;
H_rand = @gamrnd_func;
end
% ------------------------------------------------------------------
% function starts here
% ------------------------------------------------------------------
% create random set of partitions
r_parts = rand([1 length(init_parts)-1 ]) .* ( init_parts(2:end) - init_parts(1:end-1)) + init_parts(1:end-1);
P = length(r_parts);
mean_weights = zeros(N_Iter, P + 1);
% theretical mean from CDF
p = H_cdf(r_parts);
cdf_regions = [p 1] - [0 p];
theory_mean = cdf_regions;
% theretical variances
theory_variance = theory_mean .* (ones(1, P+1) - theory_mean) / (alpha + 1);
for i = 1:N_Iter
% draw random partitions
plot ( repmat(r_parts, [ 2 1]), [zeros(1,P); ones(1,P)*max_p], 'LineWidth',2,'color',[1 0 0]);
hold on;
% plot f(theta)
x_data = x_range(1):0.01:x_range(2);
plot(x_data,H_pdf(x_data),'.');
hold on;
% sample G ~ DP (alpha, H)
[sticks_weights, thetas] = Stick_breaking_process(alpha, K, H_rand);
num_samples = length(thetas);
% draw G
lineX = repmat(thetas, [1 2]);
lineY = [zeros([1 num_samples]); sticks_weights' ];
plot ( lineX', lineY, 'LineWidth',1,'color',[0 0 1]);
hold on;
for j=1:P+1
if j == 1
w_region = find(thetas < r_parts(j));
elseif j == P+1
w_region = find(thetas > r_parts(j-1));
else
w_region = find(thetas > r_parts(j-1) & thetas < r_parts(j) );
end
if ~isempty(w_region)
mean_weights(i,j) = sum(sticks_weights(w_region));
else
mean_weights(i,j) = 0;
end
end
if ~isempty(w_region)
mean_weights(i,j) = sum(sticks_weights(w_region));
else
mean_weights(i,j) = 0;
end
emperical_mean = mean(mean_weights(1:i,:),1);
if i > 1
emperical_variance = var(mean_weights(1:i,:),1);
end
for j=1:P+1
if j == 1
display_coord = (x_range(1)+ r_parts(j) )/2;
elseif j == P+1
display_coord = (r_parts(j-1) + x_range(2) )/2;
else
display_coord = (r_parts(j-1) + r_parts(j) )/2;
end
text( display_coord, max_p-0.1, num2str(theory_mean(j) ,'%.3f'), 'color', [1 0 0]);
text( display_coord, max_p-0.2, num2str(emperical_mean(j) ,'%.3f'), 'color', [1 0 0]);
text( display_coord, max_p-0.3, num2str(theory_variance(j) ,'%.3f'), 'color', [1 0 1]);
if i > 1
text( display_coord, max_p-0.4, num2str(emperical_variance(j) ,'%.3f'),'color', [1 0 1]);
else
text( display_coord, max_p-0.4, num2str(0 ,'%.3f'),'color', [1 0 1]);
end
end
set(gca,'YTick',[0.5:0.1:max_p]);
set(gca,'YTick',[max_p-0.4:0.1:max_p-0.1]);
set(gca,'YTickLabel',[ 'emperical var '; 'theoretic var '; 'emperical mean'; 'theoretic mean']);
hold on;
axis([x_range(1) x_range(2) 0 max_p]);
waitforbuttonpress;
hold off;
%sum(mean_weights(i,:))
end
% emperical mean
emperical_mean = mean(mean_weights);
display(emperical_mean);
% emperical variances
emperical_variance = var(mean_weights);
display(emperical_variance);
end
function probs = gampdf_func (x_data)
a = 2; b = 2;
probs = gampdf(x_data,a,b);
end
function probs = gamcdf_func (x_data)
a = 2; b = 2;
probs = gamcdf(x_data,a,b);
end
function x = gamrnd_func(N)
a = 2; b = 2;
x = gamrnd(a,b,[N 1]);
end
function x = nrand_func(N)
x = randn([N 1]);
end
% stick-breaking process
% when H_rand is a (continous) density function handel, it draws a sample,
% otherwise, it performs a multinomial distribution draw
function [sticks_weights, thetas] = Stick_breaking_process(alpha, N, H_rand)
raw_sticks = betarnd(1,alpha,[N 1]);
sticks_weights = zeros(N,1);
sticks_weights(1) = raw_sticks(1);
for i = 2:N
sticks_weights(i) = raw_sticks(i) * (1 - sum(sticks_weights(1:i-1)));
end
% sampling thetas
if isa(H_rand, 'function_handle')
thetas = H_rand(N);
else
for t=1:N
thetas(t) = find(rand < cumsum(H_rand),1);
end
h =1;
end
end