From 8070dc963753142bb86c4ed698d91fd623ed28e7 Mon Sep 17 00:00:00 2001 From: ziejd2 Date: Thu, 28 Sep 2017 15:04:40 -0500 Subject: BNW using Octave instead of Matlab. This version of BNW should perform the same as the original version. The only difference is that it uses Octave instead of Matlab when running BayesNet Toolbox during parameter learning. I am calling this BNW_1.02. It can be accessed at: compbio.uthsc.edu/BNW_1.02 --- sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m | 290 ++++++++++++++++++++++ 1 file changed, 290 insertions(+) create mode 100644 sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m (limited to 'sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m') diff --git a/sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m b/sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m new file mode 100644 index 00000000..4cff44d6 --- /dev/null +++ b/sourcecodes/bnt-master/SLP/misc/gener_NMAR_data.m @@ -0,0 +1,290 @@ +function [data, comp_data, bnet_miss] = gener_NMAR_dataset(bnet_orig, m, bnet_miss, upd) +% [NMAR_data] = gener_NMAR_data(bnet_miss, length_of_dataset) +% +% this function takes in input a bnet that can be used to generate NMAR data. +% this bnet bnet_miss can be creating by the function gener_NMAR_bnet. +% +% - comp_data (array) is a dataset that was generate by the bnet_orig that you enter in gener_NMAR_bnet +% - NMAR_data (cell array) is the dataset compdata that was emptyied by the NMAR process encodes in bnet_miss +% +% optional : +% - bnet_miss : an old bnet_miss built by this function +% - upd==1 if you want to update the bnet_miss +% +% [data, comp_data, bnet_miss] = gener_NMAR_dataset(bnet_orig, m, bnet_miss, upd); +% +% version 0.5 : june 8th 2005, olivier.francois@insa-rouen.fr +% +% TO DO : +% - allow the combinaison of a missing state of one variable and another state of another variable to have influence +% - allow the introduction of new nodes and specify which nodes it influence and which nodes has influence on it (it will also satisfy the first task then) +% + + +% INIT +N = size(bnet_orig.dag,2); +if mod(N,2)~=0, error('The number of nodes must be even'); end +if nargin<4, upd =0; end + +% fisrt rules +if nargin<3, + l1=[]; l2=[]; lp=[]; b=-1; + while ~(b==0 | b==1), b = input('Would you like to make a node missing when another one is missing (1 for yes, 0 for no) ? '); end +else + b=-1; + if upd, while ~(b==0 | b==1), b = input('Would you like to add rules (1 for yes, 0 for no) ? '); end + l1=bnet_miss.list{1}; + l2=bnet_miss.list{2}; + lp=bnet_miss.list{3}; + else + l1=[]; l2=[]; lp=[]; b=-1; + end +end + +while b==1, + n = input('The firts node ? '); + s = input('The node that have to be missing when this one is missing ? '); + p = input('The probability of the second node to be missing ? '); + l1 = [l1, n]; l2 = [l2, s]; lp=[lp, p]; + b=-1; + while ~(b==0 | b==1), b = input('Another one (1 for yes, 0 for no) ? '); end +end +bb = length(l1); + + +if nargin>=3, + bnet_miss = gener_NMAR_bnet(upd, bnet_orig, bnet_miss); +else + bnet_miss = gener_NMAR_bnet(1, bnet_orig); + bnet_miss.list={l1; l2; lp}; +%%%%%%%%%%% SAVING FILE + ss = 1; + if nargin == 2 | upd==1, ss = input('Would you like to save the bnet of the NMAR process you have made (1 for yes) ? '); end + if ss == 1, + ddd = datestr(now); + ddd([12 15 18])='-' ; + fnout=['NMAR-bnet-' ddd '.mat']; + eval(['save ' fnout ' bnet_miss']); + fprintf(' The bnet for NMAR process was saved as : %s\n',fnout); + end +end %if nargin + +% Generation of a complete dataset +if N>9 & m>2000, disp(' ! It could take a long time...'); end +data = cell(N,m); +for l = 1:m, data(:,l) = sample_bnet(bnet_orig); end +disp('Complete data have been creating.'); + +% Generation of a NMAR dataset +miss_array = cell(2*N,m); +vide = cell(1,N); l= 1; +while l <= m, + ev(1:N) = data(:,l); ev(N+1:2*N) = vide; + miss_array(:,l) = sample_bnet(bnet_miss, 'evidence', ev); + % apply simple rule of missingness + ev2 = cell2mat(miss_array(N+1:2*N, l)); + if bb, + missl1 = myintersect(find(ev2==2), l1); + if ~isempty(missl1), + for i=1:length(l1), + if ev2(l1(i))==2, if rand1, p = input('Base probability of a value to be missing ? '); end + for i=1:N + fam = find(dag_miss(:,N+i)==1)'; + semisize = prod(ns_miss(fam)); % as node N+i is binary to say i is present or missing + CPT = zeros(1,2*semisize); + CPT(1:semisize) = 1-p; + CPT(semisize+1:2*semisize) = p; + bnet_miss.CPD{N+i} = tabular_CPD (bnet_miss, N+i, CPT); + end +end +end + +b=-1; +while ~(b==0 | b==1), b = input('Would you like to change a probability of a node to be missing (1 for yes, 0 for no) ? '); end +if b, disp(' BE CAREFULL !! New rules can overwrite old ones partialy or fully !! So the order of entries is important'); end + +%%%%%%%%%%%% Update CPT with NMAR process +while b + fprintf('Nodes are from 1 to %d. ',N); + i=0; + while i<1 | i>N | round(i)~=i, i = input('Which node ? '); end + fam = find(dag(:,i)==1)'; + fam_miss = find(dag_miss(:,N+i)==1)'; + cas = -ones(1, length(fam_miss)+1); + familly = [fam_miss, N+i]; + fprintf('States are from 1 to %d (-1 for any states, -2 to cancel). For which state of the variable %d ?', ns(i), i); + state=-3; + while state<-2 | state>ns(i) | round(state)~=state | state==0, state = input(' ');end + if isempty(fam), + if state==-1, + p=-1; + while p<0 | p>1, p = input(' - A priori probability for this node to be missing ? ');end + semisize = prod(ns_miss(fam_miss)); + CPT = zeros(1,2*semisize); + CPT(1:semisize) = 1-p; + CPT(semisize+1:2*semisize) = p; + bnet_miss.CPD{N+i} = tabular_CPD (bnet_miss, N+i, CPT); + elseif state~=-2 + cas = state; + CPT = CPT_from_bnet(bnet_miss); + CPT = CPT{N+i}; + p=-1; + while p<0 | p>1, p = input(' - A priori probability for this node to be missing in this state ? ');end + ind = subv2ind(ns_miss(familly),[cas, 1]); + CPT(ind)=1-p; + ind = subv2ind(ns_miss(familly),[cas, 2]); + CPT(ind)=p; + bnet_miss.CPD{N+i} = tabular_CPD (bnet_miss, N+i, CPT); + end + else + if state>-2, + siz=length(cas); + place = find(fam_miss==i); + cas(place) = state; + + for k = fam, + state=-3; + fprintf(' - For the parent named %d, states are from 1 to %d (-1 for any states of this parent). ',k, ns(k)); + while state<0 | state>ns(k) | round(state)~=state, state = input('Which state ? ');end + %if state==0, + % place = find(fam_miss==(fam_miss(k)+N)); + % cas(place) = 2; % Missing + %elseif state==-2, % a changer ??? + % disp(' This case is buggy, taking missing state instand to minimise influence.'); + % place = find(fam_miss==(fam_miss(k)+N)); + % cas(place) = 2; + %elseif state~=0 & state~=-2, + place = find(fam_miss==(fam_miss(k))); + cas(place) = state; % Present + if state~=-1; place = find(fam_miss==(fam_miss(k)+N)); cas(place) = 1; end + %end + end + + p=-1; + while p<0 | p>1, p = input('Probability in this case of the value to be missing ? ');end + CPT = CPT_from_bnet(bnet_miss); + CPT = CPT{N+i}; + + cas(end) = 1; % i is present + subcas_names = find(cas==-1); + if isempty(subcas_names), + ind = subv2ind(ns_miss(familly),cas); + CPT(ind) = 1-p; + else + subcas = ones(1, length(subcas_names)); + continu = 1; + while continu + cas(subcas_names) = subcas; + ind = subv2ind(ns_miss(familly),cas); + CPT(ind) = 1-p; + [subcas, continu] = next_case(subcas, ns_miss(familly(subcas_names))); + end + end + + cas(end)=2; % i is missing + if isempty(subcas_names), + ind = subv2ind(ns_miss(familly),cas); + CPT(ind) = p; + else + subcas = ones(1, length(subcas_names)); + continu = 1; + while continu + cas(subcas_names) = subcas; + ind = subv2ind(ns_miss(familly),cas); + CPT(ind) = p; + [subcas, continu] = next_case(subcas, ns_miss(familly(subcas_names))); + end + end + mass = sum(CPT, length(size(CPT))); + while length(size(mass))>2, mass = prod(mass, length(size(mass))); end + mass = prod(prod(mass)); + if mass~=1, disp('not a proba...'); end + bnet_miss.CPD{N+i} = tabular_CPD (bnet_miss, N+i, CPT); + end %if state~=-2 for the node + end %if isempty(fam), + b=-1; + while ~(b==0 | b==1), b = input('Would you like to change a probability of a node to be missing (1 for yes, 0 for no) ? ');end +end + -- cgit 1.4.1