Mercurial > hg > camir-aes2014
comparison toolboxes/FullBNT-1.0.7/bnt/CPDs/@tabular_CPD/Old/tabular_CPD.m @ 0:e9a9cd732c1e tip
first hg version after svn
author | wolffd |
---|---|
date | Tue, 10 Feb 2015 15:05:51 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:e9a9cd732c1e |
---|---|
1 function CPD = tabular_CPD(bnet, self, varargin) | |
2 % TABULAR_CPD Make a multinomial conditional prob. distrib. (CPT) | |
3 % | |
4 % CPD = tabular_CPD(bnet, node) creates a random CPT. | |
5 % | |
6 % The following arguments can be specified [default in brackets] | |
7 % | |
8 % CPT - specifies the params ['rnd'] | |
9 % - T means use table T; it will be reshaped to the size of node's family. | |
10 % - 'rnd' creates rnd params (drawn from uniform) | |
11 % - 'unif' creates a uniform distribution | |
12 % - 'leftright' only transitions from i to i/i+1 are allowed, for each non-self parent context. | |
13 % The non-self parents are all parents except oldself. | |
14 % selfprob - The prob of transition from i to i if CPT = 'leftright' [0.1] | |
15 % old_self - id of the node corresponding to self in the previous slice [self-ss] | |
16 % adjustable - 0 means don't adjust the parameters during learning [1] | |
17 % prior_type - defines type of prior ['none'] | |
18 % - 'none' means do ML estimation | |
19 % - 'dirichlet' means add pseudo-counts to every cell | |
20 % - 'entropic' means use a prior P(theta) propto exp(-H(theta)) (see Brand) | |
21 % dirichlet_weight - equivalent sample size (ess) of the dirichlet prior [1] | |
22 % dirichlet_type - defines the type of Dirichlet prior ['BDeu'] | |
23 % - 'unif' means put dirichlet_weight in every cell | |
24 % - 'BDeu' means we put 'dirichlet_weight/(r q)' in every cell | |
25 % where r = self_sz and q = prod(parent_sz) (see Heckerman) | |
26 % trim - 1 means trim redundant params (rows in CPT) when using entropic prior [0] | |
27 % | |
28 % e.g., tabular_CPD(bnet, i, 'CPT', T) | |
29 % e.g., tabular_CPD(bnet, i, 'CPT', 'unif', 'dirichlet_weight', 2, 'dirichlet_type', 'unif') | |
30 % | |
31 % REFERENCES | |
32 % M. Brand - "Structure learning in conditional probability models via an entropic prior | |
33 % and parameter extinction", Neural Computation 11 (1999): 1155--1182 | |
34 % M. Brand - "Pattern discovery via entropy minimization" [covers annealing] | |
35 % AI & Statistics 1999. Equation numbers refer to this paper, which is available from | |
36 % www.merl.com/reports/docs/TR98-21.pdf | |
37 % D. Heckerman, D. Geiger and M. Chickering, | |
38 % "Learning Bayesian networks: the combination of knowledge and statistical data", | |
39 % Microsoft Research Tech Report, 1994 | |
40 | |
41 | |
42 if nargin==0 | |
43 % This occurs if we are trying to load an object from a file. | |
44 CPD = init_fields; | |
45 CPD = class(CPD, 'tabular_CPD', discrete_CPD(0, [])); | |
46 return; | |
47 elseif isa(bnet, 'tabular_CPD') | |
48 % This might occur if we are copying an object. | |
49 CPD = bnet; | |
50 return; | |
51 end | |
52 CPD = init_fields; | |
53 | |
54 ns = bnet.node_sizes; | |
55 ps = parents(bnet.dag, self); | |
56 fam_sz = ns([ps self]); | |
57 CPD.sizes = fam_sz; | |
58 CPD.leftright = 0; | |
59 | |
60 % set defaults | |
61 CPD.CPT = mk_stochastic(myrand(fam_sz)); | |
62 CPD.adjustable = 1; | |
63 CPD.prior_type = 'none'; | |
64 dirichlet_type = 'BDeu'; | |
65 dirichlet_weight = 1; | |
66 CPD.trim = 0; | |
67 selfprob = 0.1; | |
68 | |
69 % extract optional args | |
70 args = varargin; | |
71 % check for old syntax CPD(bnet, i, CPT) as opposed to CPD(bnet, i, 'CPT', CPT) | |
72 if ~isempty(args) & ~isstr(args{1}) | |
73 CPD.CPT = myreshape(args{1}, fam_sz); | |
74 args = []; | |
75 end | |
76 | |
77 % if old_self is specified, read in the value before CPT is created | |
78 old_self = []; | |
79 for i=1:2:length(args) | |
80 switch args{i}, | |
81 case 'old_self', old_self = args{i+1}; | |
82 end | |
83 end | |
84 | |
85 for i=1:2:length(args) | |
86 switch args{i}, | |
87 case 'CPT', | |
88 T = args{i+1}; | |
89 if ischar(T) | |
90 switch T | |
91 case 'unif', CPD.CPT = mk_stochastic(myones(fam_sz)); | |
92 case 'rnd', CPD.CPT = mk_stochastic(myrand(fam_sz)); | |
93 case 'leftright', | |
94 % we just initialise the CPT to leftright - this structure will | |
95 % be maintained by EM, assuming we don't use a prior... | |
96 CPD.leftright = 1; | |
97 if isempty(old_self) % we assume the network is a DBN | |
98 ss = bnet.nnodes_per_slice; | |
99 old_self = self-ss; | |
100 end | |
101 other_ps = mysetdiff(ps, old_self); | |
102 Qps = prod(ns(other_ps)); | |
103 Q = ns(self); | |
104 p = selfprob; | |
105 LR = mk_leftright_transmat(Q, p); | |
106 transprob = repmat(reshape(LR, [1 Q Q]), [Qps 1 1]); % transprob(k,i,j) | |
107 transprob = permute(transprob, [2 1 3]); % now transprob(i,k,j) | |
108 CPD.CPT = myreshape(transprob, fam_sz); | |
109 otherwise, error(['invalid CPT ' T]); | |
110 end | |
111 else | |
112 CPD.CPT = myreshape(T, fam_sz); | |
113 end | |
114 | |
115 case 'prior_type', CPD.prior_type = args{i+1}; | |
116 case 'dirichlet_type', dirichlet_type = args{i+1}; | |
117 case 'dirichlet_weight', dirichlet_weight = args{i+1}; | |
118 case 'adjustable', CPD.adjustable = args{i+1}; | |
119 case 'clamped', CPD.adjustable = ~args{i+1}; | |
120 case 'trim', CPD.trim = args{i+1}; | |
121 case 'old_self', noop = 1; % already read in | |
122 otherwise, error(['invalid argument name: ' args{i}]); | |
123 end | |
124 end | |
125 | |
126 switch CPD.prior_type | |
127 case 'dirichlet', | |
128 switch dirichlet_type | |
129 case 'unif', CPD.dirichlet = dirichlet_weight * myones(fam_sz); | |
130 case 'BDeu', CPD.dirichlet = dirichlet_weight * mk_stochastic(myones(fam_sz)); | |
131 otherwise, error(['invalid dirichlet_type ' dirichlet_type]) | |
132 end | |
133 case {'entropic', 'none'} | |
134 CPD.dirichlet = []; | |
135 otherwise, error(['invalid prior_type ' prior_type]) | |
136 end | |
137 | |
138 | |
139 | |
140 % fields to do with learning | |
141 if ~CPD.adjustable | |
142 CPD.counts = []; | |
143 CPD.nparams = 0; | |
144 CPD.nsamples = []; | |
145 else | |
146 CPD.counts = zeros(size(CPD.CPT)); | |
147 psz = fam_sz(1:end-1); | |
148 ss = fam_sz(end); | |
149 if CPD.leftright | |
150 % For each of the Qps contexts, we specify Q elements on the diagoanl | |
151 CPD.nparams = Qps * Q; | |
152 else | |
153 % sum-to-1 constraint reduces the effective arity of the node by 1 | |
154 CPD.nparams = prod([psz ss-1]); | |
155 end | |
156 CPD.nsamples = 0; | |
157 end | |
158 | |
159 fam_sz = CPD.sizes; | |
160 psz = prod(fam_sz(1:end-1)); | |
161 ssz = fam_sz(end); | |
162 CPD.trimmed_trans = zeros(psz, ssz); % must declare before reading | |
163 | |
164 CPD = class(CPD, 'tabular_CPD', discrete_CPD(~CPD.adjustable, fam_sz)); | |
165 | |
166 | |
167 %%%%%%%%%%% | |
168 | |
169 function CPD = init_fields() | |
170 % This ensures we define the fields in the same order | |
171 % no matter whether we load an object from a file, | |
172 % or create it from scratch. (Matlab requires this.) | |
173 | |
174 CPD.CPT = []; | |
175 CPD.sizes = []; | |
176 CPD.prior_type = []; | |
177 CPD.dirichlet = []; | |
178 CPD.adjustable = []; | |
179 CPD.counts = []; | |
180 CPD.nparams = []; | |
181 CPD.nsamples = []; | |
182 CPD.trim = []; | |
183 CPD.trimmed_trans = []; | |
184 CPD.leftright = []; | |
185 | |
186 |