Add files via upload

Adding the *.m files used in parameter learning.
author: ziejd2 2017-09-14 16:17:47 -0500
committer: GitHub 2017-09-14 16:17:47 -0500
commit: 7cc31810d53176e805532b2789955f4eedbce6bb (patch)
tree: 82924642070d871f753ee41c0f3e363ff8f380da
parent: 6882395afdadf4e982b25b5215071a0932730950 (diff)
download: BNW-7cc31810d53176e805532b2789955f4eedbce6bb.tar.gz
13 files changed, 1265 insertions, 0 deletions
diff --git a/BNW_parameter_learning/Predictmultiple.m b/BNW_parameter_learning/Predictmultiple.m
new file mode 100644
index 00000000..c14b516c
--- /dev/null
+++ b/BNW_parameter_learning/Predictmultiple.m
@@ -0,0 +1,57 @@
+function Predictmultiple(pre)
+
+dfile=strcat(pre,'structure_input.txt');
+sfile=dfile;
+dfile=strcat(pre,'continuous_input.txt');
+nnodefile=strcat(pre,'nnode.txt');
+fnnode = fopen(nnodefile,'r');
+nnodes = fscanf(fnnode,'%d');
+
+
+
+%nnodes=5;
+Std_flag=true;
+[labels,cases,bnet]=readInput(dfile,sfile,nnodes,Std_flag);
+
+
+%name
+%labels
+%map
+
+[bnet]=parameterLearning(bnet,cases);
+%[predict_mean,predict_sd,q_sq]=looCrossValid(bnet,cases);
+fvarfile=strcat(pre,'var.txt');
+
+fvar = fopen(fvarfile,'r');
+                            
+select_var_new = fscanf(fvar,'%d');
+
+fvardfile=strcat(pre,'vardata.txt');
+
+fvard = fopen(fvardfile,'r');
+
+select_var_data_new = fscanf(fvard,'%f');
+
+filename=strcat(pre,'net_figure_new.txt');
+
+drawFigureM(nnodes,bnet,labels,filename,cases,select_var_new,select_var_data_new);
+
+%quit force;
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{1}=2;
+%[engine,loglik]=enter_evidence(engine,evidence)
+%marginal_nodes(engine,1)
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{2}=0.6;
+%evidence{1}=[];
+%[engine,loglik]=enter_evidence(engine,evidence);
+%marginal_nodes(engine,3);
+%marginal_nodes(engine,4);
+%marginal_nodes(engine,5);
+end
\ No newline at end of file
diff --git a/BNW_parameter_learning/Predictmultipleintrvention.m b/BNW_parameter_learning/Predictmultipleintrvention.m
new file mode 100644
index 00000000..4675b846
--- /dev/null
+++ b/BNW_parameter_learning/Predictmultipleintrvention.m
@@ -0,0 +1,73 @@
+function Predictmultipleintrvention(pre)
+dfile=strcat(pre,'structure_input.txt');
+sfile=dfile;
+dfile=strcat(pre,'continuous_input.txt');
+nnodefile=strcat(pre,'nnode.txt');
+
+fnnode = fopen(nnodefile,'r');
+nnodes = fscanf(fnnode,'%d');
+
+fvarnamefile=strcat(pre,'varname.txt');
+
+varfile = fopen(fvarnamefile,'r');
+
+%nnodes=5;
+Std_flag=true;
+[labels,cases,bnet]=readInput(dfile,sfile,nnodes,Std_flag);
+[bnet]=parameterLearning(bnet,cases);
+
+%[predict_mean,predict_sd,q_sq]=looCrossValid(bnet,cases);
+
+
+fvarfile=strcat(pre,'var.txt');
+fvar = fopen(fvarfile,'r');
+                            
+select_var_new = fscanf(fvar,'%d');
+
+nm = numel(select_var_new);
+
+varlabels = cell(1,nm);
+varbuffer = fgetl(varfile);    %get header line as a string
+for j=1:nm
+    [varnext,varbuffer] = strtok(varbuffer);
+    varlabels{j} = varnext;
+    for i=1:nnodes    
+        if strcmp(varlabels{j},labels{i})
+            select_var_new(j)=i;
+        end
+     end    
+    
+end
+
+
+
+
+fvardfile=strcat(pre,'vardata.txt');
+
+fvard = fopen(fvardfile,'r');
+
+select_var_data_new = fscanf(fvard,'%f');
+
+filename=strcat(pre,'net_figure_new.txt');
+
+drawFigureM(nnodes,bnet,labels,filename,cases,select_var_new,select_var_data_new);
+
+%quit force;
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{1}=2;
+%[engine,loglik]=enter_evidence(engine,evidence)
+%marginal_nodes(engine,1)
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{2}=0.6;
+%evidence{1}=[];
+%[engine,loglik]=enter_evidence(engine,evidence);
+%marginal_nodes(engine,3);
+%marginal_nodes(engine,4);
+%marginal_nodes(engine,5);
+end
\ No newline at end of file
diff --git a/BNW_parameter_learning/checkDiscreteNodes.m b/BNW_parameter_learning/checkDiscreteNodes.m
new file mode 100644
index 00000000..7b2a695f
--- /dev/null
+++ b/BNW_parameter_learning/checkDiscreteNodes.m
@@ -0,0 +1,37 @@
+function [ ] = checkDiscreteNodes( bnet, cases)
+    %checkDiscreteNodes Checks if states of discrete nodes are be integers from 1 to M
+    %   where M is the number of states of the node.  (M should be the same as
+    %   node_sizes in the bnet).
+    % 
+    %Input:
+    %   bnet: BNT bnet
+    %   cases: cell array of data
+    %
+%
+node_sizes = bnet.node_sizes;
+dnodes = bnet.dnodes;
+ndisc = size(dnodes,2);
+ncases = size(cases,2);
+
+%check to see that all data for discrete nodes are integers
+for i = 1:ndisc
+    inode = dnodes(i);
+    data = cases(inode,:);
+    isize = node_sizes(inode);
+    states = zeros(1,isize);
+    for j = 1:isize
+        states(j) = j;
+    end
+    for j = 1:ncases
+        k = int64(data{j});
+        if ~any(k==states)
+            error(['Discrete nodes must be integers from 1 to the number of states']);
+        end
+    end
+end
+
+    
+end
+
+
+
diff --git a/BNW_parameter_learning/checkStructure.m b/BNW_parameter_learning/checkStructure.m
new file mode 100644
index 00000000..fc72e4b7
--- /dev/null
+++ b/BNW_parameter_learning/checkStructure.m
@@ -0,0 +1,78 @@
+function [ labels, cases, dag, node_sizes, ord_flag ] = checkStructure(labels, cases, dag, node_sizes)
+    %checkStructure Check to see if nodes are sorted correctly.  They must be
+    %   in topological order (i.e., parents before children) before parameter
+    %   learning can take place.
+    %
+    %Input and output have the same meaning.  The output has just been
+    %topologically ordered.
+    %   labels = cell array with the names of the nodes.
+    %   cases = cell array with the data.
+    %   dag = matrix with the strucutre of the network.
+    %   node_sizes = vector with the size of each node.
+
+%make connections array
+%count how big you need the connections array to be
+nnodes = size(dag,1);
+narcs = 0;
+for i = 1:nnodes
+    for j = 1:nnodes
+        if dag(i,j) == 1
+            narcs = narcs + 1;
+        end
+    end
+end
+%fill connections array with label names
+connections = cell(narcs,2);
+ncount = 0;
+for i = 1:nnodes
+    for j = 1:nnodes
+        if dag(i,j) == 1
+            ncount = ncount + 1;
+            connections{ncount,1} = labels{i};
+            connections{ncount,2} = labels{j};
+        end
+    end
+end
+
+%get topologically sorted dag and labels
+[new_dag, new_labels] = mk_adj_mat(connections, labels, 1);
+
+%check to see if order changed
+ord_flag = 0;
+for i = 1:nnodes
+    if ~strcmp(new_labels{i},labels{i})
+        ord_flag = 1;
+    end
+end
+
+if ord_flag
+    %get new ordering of nodes
+    order = cell(1,nnodes);
+    for i = 1:nnodes
+        for j = 1:nnodes
+            if strcmp(new_labels{j},labels{i})
+                order{i} = j;
+            end
+        end
+    end
+
+    %reorder cases and node_sizes
+    new_cases = cell(size(cases));
+    for i = 1:nnodes
+        new_cases(order{i},:) = cases(i,:);
+    end
+    new_node_sizes = zeros(1,nnodes);
+    for i = 1:nnodes
+        new_node_sizes(order{i}) = node_sizes(i);
+    end
+
+
+    dag = new_dag;
+    cases = new_cases;
+    node_sizes = new_node_sizes;
+    labels = new_labels;    
+end
+
+end
+%end checkStructure.m
+
diff --git a/BNW_parameter_learning/drawFigure.m b/BNW_parameter_learning/drawFigure.m
new file mode 100644
index 00000000..a5de0f2d
--- /dev/null
+++ b/BNW_parameter_learning/drawFigure.m
@@ -0,0 +1,383 @@
+function [] = drawFigure(nnodes,bnet,labels,filename,cases,selectvar,selectdata)
+%drawFigure writes the parameters and data that are needed to draw the
+%structure of a Bayesian network.
+
+if nargin < 6,
+    drawFigureNoEv(nnodes,bnet,labels,filename,cases);
+else
+    drawFigureEv(nnodes,bnet,labels,filename,cases,selectvar,selectdata);
+end;
+
+end
+
+
+
+function [] = drawFigureEv(nnodes,bnet,labels,filename,cases,selectvar,selectdata)
+%Function to use if there is no entered evidence. 
+%         
+%
+%Before each printed line, I will have a line that starts with %%%
+% that describes what will be on that line
+
+%Create an empty evidence cell array.
+
+%val=cases;
+%for i = 1:nnodes
+% val(i,1)=val(i,2);
+
+%end
+
+A=cell2mat(cases');
+Amax=max(A);
+Amin=min(A);
+
+
+evidence = cell(1,nnodes);
+engine = jtree_inf_engine(bnet);
+
+evidence{selectvar}=selectdata;
+
+[engine,loglik]=enter_evidence(engine,evidence);
+
+%Open the file, and write the nodes to a file.
+fileID = fopen(filename,'w');
+
+%%%%Evidence node
+fprintf(fileID,'%i\n',selectvar);
+%%% The number of nodes
+fprintf(fileID,'%i\n',nnodes);
+%Get canvas size
+labels_temp = cellstr(labels);
+[x,y] = make_layout(bnet.dag);
+
+x = x - min(x);
+y = 1 - y;
+y = y - min(y);
+
+[x_dim,y_dim] = canvasSize(nnodes,x,y);
+
+%%% The dimensions of the canvas for the javascript code
+fprintf(fileID,'%i\t%i\t\n',x_dim,y_dim)
+
+x = x*x_dim;
+y = y*y_dim;
+for i = 1:nnodes,
+%%% The name and X- and Y-positions of each node
+    fprintf(fileID,'%s\t%i\t%i\n',labels{i},round(x(i)),round(y(i)));
+end
+
+%Get the number of parents and children for each node.
+num_par = zeros(1,nnodes);
+%For parents, sum down columns
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(j,i) == 1,
+            num_par(i) = num_par(i) + 1;
+        end
+    end
+end
+num_child = zeros(1,nnodes);
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(i,j) == 1,
+            num_child(i) = num_child(i) + 1;
+        end
+    end
+end
+
+
+for i = 1:nnodes,
+    %%% The name and type of each node (1=continuous, the number of states
+    %%% if it is discrete
+    fprintf(fileID,'%s\t%i\n',labels{i},bnet.node_sizes(i));
+    %%% The size of the node, I am going to keep them 
+    %%% 250(width) by 150(height) for now
+    %Could modify this to change the width based on the length of the node
+    %name
+    fprintf(fileID,'%i\t%i\n',250,150);
+    %%% The number of parents of the node, and the parents
+    if num_par(i) == 0;
+        %%% If no parents:
+        fprintf(fileID,'%i\n',num_par(i));
+    else
+        parents = zeros(1,num_par(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(j,i) == 1,
+             parents(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_par(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_par(i),parents(1,:));
+    end
+    
+    
+    %%% The number of children of the node, and the children
+    if num_child(i) == 0;
+        %%% If no children:
+        fprintf(fileID,'%i\n',num_child(i));
+    else
+        children = zeros(1,num_child(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(i,j) == 1,
+             children(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_child(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_child(i),children(1,:));
+    end
+    
+    predict = marginal_nodes(engine,i);
+    if isempty(evidence{i})
+      if bnet.node_sizes(i) ~= 1,
+        for j = 1:bnet.node_sizes(i),
+            %%%For discrete nodes, the state and the percent of that state
+            fprintf(fileID,'%i\t%6.4f\n',j,predict.T(j));
+        end;
+      else
+
+        [x_vals,y_vals] = calcGaussian(predict.mu,predict.Sigma,Amax(i),Amin(i));
+        %%%For continuous nodes, print x and the pdf of a normal curve.
+        for j = 1:101,
+            fprintf(fileID,'%6.4f\t%6.4f\n',x_vals(j,1),y_vals(j,1));
+        end;
+      end;
+    else
+      fprintf(fileID,'%6.4f\t%6.4f\n',selectdata,1);
+    end
+    
+end
+%fprintf(fileID,'%s\t %\n',labels_temp{:});
+
+
+fclose(fileID);
+
+end
+
+
+
+
+
+
+function [] = drawFigureNoEv(nnodes,bnet,labels,filename,cases)
+%Function to use if there is no entered evidence. 
+%         
+%
+%Before each printed line, I will have a line that starts with %%%
+% that describes what will be on that line
+A=cell2mat(cases');
+Amax=max(A);
+Amin=min(A);
+
+%Create an empty evidence cell array.
+evidence = cell(1,nnodes);
+engine = jtree_inf_engine(bnet);
+[engine,loglik] = enter_evidence(engine,evidence);
+
+%Open the file, and write the nodes to a file.
+fileID = fopen(filename,'w');
+%%% The number of nodes
+fprintf(fileID,'%i\n',nnodes);
+
+%Get canvas size
+
+labels_temp = cellstr(labels);
+[x,y] = make_layout(bnet.dag);
+%[x,y] = layout_dag(bnet.dag);
+
+
+x = x - min(x);
+y = 1 - y;
+y = y - min(y);
+
+[x_dim,y_dim] = canvasSize(nnodes,x,y);
+
+%%% The dimensions of the canvas for the javascript code
+fprintf(fileID,'%i\t%i\t\n',x_dim,y_dim)
+
+x = x*x_dim;
+y = y*y_dim;
+for i = 1:nnodes,
+%%% The name and X- and Y-positions of each node
+    fprintf(fileID,'%s\t%i\t%i\n',labels{i},round(x(i)),round(y(i)));
+end
+
+%Get the number of parents and children for each node.
+num_par = zeros(1,nnodes);
+%For parents, sum down columns
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(j,i) == 1,
+            num_par(i) = num_par(i) + 1;
+        end
+    end
+end
+num_child = zeros(1,nnodes);
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(i,j) == 1,
+            num_child(i) = num_child(i) + 1;
+        end
+    end
+end
+
+
+for i = 1:nnodes,
+    %%% The name and type of each node (1=continuous, the number of states
+    %%% if it is discrete
+    fprintf(fileID,'%s\t%i\n',labels{i},bnet.node_sizes(i));
+    %%% The size of the node, I am going to keep them 
+    %%% 250(width) by 150(height) for now
+    %Could modify this to change the width based on the length of the node
+    %name
+    fprintf(fileID,'%i\t%i\n',250,150);
+    %%% The number of parents of the node, and the parents
+    if num_par(i) == 0;
+        %%% If no parents:
+        fprintf(fileID,'%i\n',num_par(i));
+    else
+        parents = zeros(1,num_par(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(j,i) == 1,
+             parents(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_par(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_par(i),parents(1,:));
+    end
+    
+    
+    %%% The number of children of the node, and the children
+    if num_child(i) == 0;
+        %%% If no children:
+        fprintf(fileID,'%i\n',num_child(i));
+    else
+        children = zeros(1,num_child(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(i,j) == 1,
+             children(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_child(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_child(i),children(1,:));
+    end
+    
+    predict = marginal_nodes(engine,i);
+    if bnet.node_sizes(i) ~= 1,
+        for j = 1:bnet.node_sizes(i),
+            %%%For discrete nodes, the state and the percent of that state
+            fprintf(fileID,'%i\t%6.4f\n',j,predict.T(j));
+        end;
+    else
+        %cases(i)
+       % MAX(cases(i))
+       % MIN(cases(i))
+        [x_vals,y_vals] = calcGaussian(predict.mu,predict.Sigma,Amax(i),Amin(i));
+        %%%For continuous nodes, print x and the pdf of a normal curve.
+        for j = 1:101,
+            fprintf(fileID,'%6.4f\t%6.4f\n',x_vals(j,1),y_vals(j,1));
+        end;
+    end;
+end
+%fprintf(fileID,'%s\t %\n',labels_temp{:});
+
+
+fclose(fileID);
+
+end
+
+
+function [x_dim, y_dim] = canvasSize(nnodes,x,y)
+%canvasSize Function to calculate the size of the canvas to
+%           build the network structure
+
+
+%I am going to assume that the node size will be
+% height = 150, width = 250
+% so there will be a node spacing of 
+% 200 (in y-dim) and 300 (in x-dim).
+y_space = 200;
+x_space = 300;
+
+%Set default minimum x and y dimensions
+x_dim = 1200;
+y_dim = 1200;
+
+%get unique y values
+y_unique = unique(y);
+size_y = size(y_unique,2);
+y_dim_temp = size_y*y_space;
+
+%get the maximum nodes in any layer
+size_x = zeros(1,size_y);
+for i = 1:size_y,
+    for j = 1:nnodes,
+        if y_unique(i) == y(j),
+            size_x(1,i) = size_x(1,i) + 1;
+        end;
+    end;
+end;
+size_x = max(size_x);
+x_dim_temp = size_x*x_space;
+
+if x_dim_temp > x_dim,
+    x_dim = x_dim_temp;
+end;
+
+if y_dim_temp > y_dim,
+    y_dim = y_dim_temp;
+end;
+end
+
+function [x_vals,y_vals] = calcGaussian(mu,Sigma,maxval,minval)
+%Function to calculate 101 points of Gaussian function to use in plotting
+% Gets the probability density of the mean value and 50 evenly spaced
+% points up to 3Sigma below the mean and 50 evenly space points up to
+% 3Sigma above the mean.
+%maxval
+%minval
+x_vals = zeros(101,1);
+y_vals = zeros(101,1);
+
+%x_vals(1,1) = mu - 3*Sigma;
+x_vals(1,1) = minval - 1;
+gap=((maxval+1)-(minval - 1))/100;
+%x_vals(1,1) = 0;%mu - 3*Sigma;
+for i = 1:100,
+   % x_vals(i+1,1) = x_vals(1,1) + i*6*Sigma/100;
+    x_vals(i+1,1) = x_vals(i,1) + gap;
+ %x_vals(i+1,1) = x_vals(i,1) + 1/100;
+end
+
+for i = 1:101,
+    y_vals(i,1) = normpdf(x_vals(i,1),mu,Sigma);
+end
+
+end
diff --git a/BNW_parameter_learning/drawFigureM.m b/BNW_parameter_learning/drawFigureM.m
new file mode 100644
index 00000000..bca223ce
--- /dev/null
+++ b/BNW_parameter_learning/drawFigureM.m
@@ -0,0 +1,258 @@
+function [] = drawFigureM(nnodes,bnet,labels,filename,cases,selectvar,selectdata)
+%drawFigure writes the parameters and data that are needed to draw the
+%structure of a Bayesian network.
+
+
+%Function to use if there is no entered evidence. 
+%         
+%
+%Before each printed line, I will have a line that starts with %%%
+% that describes what will be on that line
+
+%Create an empty evidence cell array.
+
+%val=cases;
+%for i = 1:nnodes
+% val(i,1)=val(i,2);
+
+%end
+
+fileID = fopen(filename,'w');
+
+
+A=cell2mat(cases');
+Amax=max(A);
+Amin=min(A);
+
+
+evidence = cell(1,nnodes);
+engine = jtree_inf_engine(bnet);
+
+m = size(selectvar,1);
+
+
+ev_dat = zeros(1,nnodes);
+%For parents, sum down columns
+for i = 1:m,
+    di=selectvar(i,1);
+    ev_dat(di)=selectdata(i,1);   
+    
+    evidence{di}=ev_dat(di);
+    fprintf(fileID,'%i\t',di);
+end
+fprintf(fileID,'\n');
+
+%ev_dat
+
+% select_var = selectvar(1,1)
+% 
+% select_var_data = selectdata(1,1)
+% 
+% 
+% evidence{select_var}=select_var_data;
+
+[engine,loglik]=enter_evidence(engine,evidence);
+
+%Open the file, and write the nodes to a file.
+
+
+%%%%Evidence node
+
+%%% The number of nodes
+fprintf(fileID,'%i\n',nnodes);
+%Get canvas size
+labels_temp = cellstr(labels);
+[x,y] = make_layout(bnet.dag);
+
+x = x - min(x);
+y = 1 - y;
+y = y - min(y);
+
+[x_dim,y_dim] = canvasSize(nnodes,x,y);
+
+%%% The dimensions of the canvas for the javascript code
+fprintf(fileID,'%i\t%i\t\n',x_dim,y_dim)
+
+x = x*x_dim;
+y = y*y_dim;
+for i = 1:nnodes,
+%%% The name and X- and Y-positions of each node
+    fprintf(fileID,'%s\t%i\t%i\n',labels{i},round(x(i)),round(y(i)));
+end
+
+%Get the number of parents and children for each node.
+num_par = zeros(1,nnodes);
+%For parents, sum down columns
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(j,i) == 1,
+            num_par(i) = num_par(i) + 1;
+        end
+    end
+end
+num_child = zeros(1,nnodes);
+for i = 1:nnodes,
+    for j = 1:nnodes,
+        if bnet.dag(i,j) == 1,
+            num_child(i) = num_child(i) + 1;
+        end
+    end
+end
+
+
+for i = 1:nnodes,
+    %%% The name and type of each node (1=continuous, the number of states
+    %%% if it is discrete
+    fprintf(fileID,'%s\t%i\n',labels{i},bnet.node_sizes(i));
+    %%% The size of the node, I am going to keep them 
+    %%% 250(width) by 150(height) for now
+    %Could modify this to change the width based on the length of the node
+    %name
+    fprintf(fileID,'%i\t%i\n',250,150);
+    %%% The number of parents of the node, and the parents
+    if num_par(i) == 0;
+        %%% If no parents:
+        fprintf(fileID,'%i\n',num_par(i));
+    else
+        parents = zeros(1,num_par(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(j,i) == 1,
+             parents(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_par(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_par(i),parents(1,:));
+    end
+    
+    
+    %%% The number of children of the node, and the children
+    if num_child(i) == 0;
+        %%% If no children:
+        fprintf(fileID,'%i\n',num_child(i));
+    else
+        children = zeros(1,num_child(i));
+        k = 1;
+        for j = 1:nnodes,
+           if bnet.dag(i,j) == 1,
+             children(1,k) = j;
+             k = k + 1;
+           end
+        end
+        format = '%i\t';
+        for j = 1:num_child(i)-1,
+            format = strcat(format,'%i\t');
+        end
+        format = strcat(format,'%i\n');
+        %%%If there are parents:
+        fprintf(fileID,format,num_child(i),children(1,:));
+    end
+    
+    predict = marginal_nodes(engine,i);
+    if isempty(evidence{i})
+      if bnet.node_sizes(i) ~= 1,
+        for j = 1:bnet.node_sizes(i),
+            %%%For discrete nodes, the state and the percent of that state
+            fprintf(fileID,'%i\t%6.4f\n',j,predict.T(j));
+        end;
+      else
+
+        [x_vals,y_vals] = calcGaussian(predict.mu,predict.Sigma,Amax(i),Amin(i));
+        %%%For continuous nodes, print x and the pdf of a normal curve.
+        for j = 1:101,
+            fprintf(fileID,'%6.4f\t%6.4f\n',x_vals(j,1),y_vals(j,1));
+        end;
+      end;
+    else
+      fprintf(fileID,'%6.4f\t%6.4f\n',ev_dat(i),1);
+    end
+    
+end
+%fprintf(fileID,'%s\t %\n',labels_temp{:});
+
+
+fclose(fileID);
+
+end
+
+
+
+
+
+
+
+
+
+function [x_dim, y_dim] = canvasSize(nnodes,x,y)
+%canvasSize Function to calculate the size of the canvas to
+%           build the network structure
+
+
+%I am going to assume that the node size will be
+% height = 150, width = 250
+% so there will be a node spacing of 
+% 200 (in y-dim) and 300 (in x-dim).
+y_space = 200;
+x_space = 300;
+
+%Set default minimum x and y dimensions
+x_dim = 1200;
+y_dim = 1200;
+
+%get unique y values
+y_unique = unique(y);
+size_y = size(y_unique,2);
+y_dim_temp = size_y*y_space;
+
+%get the maximum nodes in any layer
+size_x = zeros(1,size_y);
+for i = 1:size_y,
+    for j = 1:nnodes,
+        if y_unique(i) == y(j),
+            size_x(1,i) = size_x(1,i) + 1;
+        end;
+    end;
+end;
+size_x = max(size_x);
+x_dim_temp = size_x*x_space;
+
+if x_dim_temp > x_dim,
+    x_dim = x_dim_temp;
+end;
+
+if y_dim_temp > y_dim,
+    y_dim = y_dim_temp;
+end;
+end
+
+function [x_vals,y_vals] = calcGaussian(mu,Sigma,maxval,minval)
+%Function to calculate 101 points of Gaussian function to use in plotting
+% Gets the probability density of the mean value and 50 evenly spaced
+% points up to 3Sigma below the mean and 50 evenly space points up to
+% 3Sigma above the mean.
+%maxval
+%minval
+x_vals = zeros(101,1);
+y_vals = zeros(101,1);
+
+%x_vals(1,1) = mu - 3*Sigma;
+x_vals(1,1) = minval - 1;
+gap=((maxval+1)-(minval - 1))/100;
+%x_vals(1,1) = 0;%mu - 3*Sigma;
+for i = 1:100,
+   % x_vals(i+1,1) = x_vals(1,1) + i*6*Sigma/100;
+    x_vals(i+1,1) = x_vals(i,1) + gap;
+ %x_vals(i+1,1) = x_vals(i,1) + 1/100;
+end
+
+for i = 1:101,
+    y_vals(i,1) = normpdf(x_vals(i,1),mu,Sigma);
+end
+
+end
diff --git a/BNW_parameter_learning/getParams.m b/BNW_parameter_learning/getParams.m
new file mode 100644
index 00000000..4840b38d
--- /dev/null
+++ b/BNW_parameter_learning/getParams.m
@@ -0,0 +1,22 @@
+function [ bnet ] = getParams( bnet, cases )
+%getParams Code to initialize CPT and do parameter learning.
+%This will be very basic for now.  I can add more options later.
+
+dnodes = bnet.dnodes;
+cnodes = bnet.cnodes;
+nnodes = size(dnodes,2)+size(cnodes,2);
+
+%make dnodes tabular_CPT
+for i = 1:size(dnodes,2)
+    bnet.CPD{dnodes(i)} = tabular_CPD(bnet,dnodes(i));
+end
+
+for i = 1:size(cnodes,2)
+    bnet.CPD{cnodes(i)} = gaussian_CPD(bnet,cnodes(i));
+end
+
+bnet = learn_params(bnet,cases);
+
+
+end
+
diff --git a/BNW_parameter_learning/parameterLearning.m b/BNW_parameter_learning/parameterLearning.m
new file mode 100644
index 00000000..d4b67c87
--- /dev/null
+++ b/BNW_parameter_learning/parameterLearning.m
@@ -0,0 +1,17 @@
+function [ bnet ] = parameterLearning( bnet,cases,engine_name )
+%parameterLearning Do parameter learning and inference
+
+%engine is an optional argument
+if nargin < 3
+    engine_name = 'jtree_inf_engine';
+end
+
+
+%First do parameter learning with all the data
+[bnet] = getParams(bnet,cases);
+
+
+
+
+end
+
diff --git a/BNW_parameter_learning/readInput.m b/BNW_parameter_learning/readInput.m
new file mode 100644
index 00000000..c2c16331
--- /dev/null
+++ b/BNW_parameter_learning/readInput.m
@@ -0,0 +1,70 @@
+function [ labels, cases, bnet, node_sizes, data,labelsold] = readInput( dfile, sfile, nnodes, std_flag )
+    %readInput is to be used when reading in a network with a known structure
+    %   
+    %Input:
+	%   dfile  = name of the file containing the data (required)
+    %   sfile = name of the file containing the structure (required)
+    %   nnodes = number of nodes in the network (required)
+    %   std_flag = flag for whether or not to standardize the data.
+    %   (optional-- Default is FALSE)
+    %
+    %   See readInputData.m and readInputStructure.m for description of the
+    %       format of the dfile and sfile, respectively. 
+    %
+    %Output:
+    %   labels = cell array with the names of the nodes.
+    %   cases = cell array with the data.
+    %   bnet = BNT bayesian network with the input structure.
+
+if nargin < 4
+    std_flag = false(1);
+end
+
+    
+% read in the file with the data
+[labelsold,node_sizes,cases, data] = readInputData(dfile,nnodes);
+
+
+
+
+% read in the file with the structure
+[dag] = readInputStructure(sfile,labelsold);
+
+
+% check the ordering of the nodes and reorder if necessary
+[labels,cases,dag,node_sizes,ord_flag] = checkStructure(labelsold,cases,dag,node_sizes);
+%draw_graph(dag,labels);
+%if ord_flag == 1
+%    fprintf(['Order of nodes was changed to agree with topological order\n'])
+%end
+%fprintf(['The structure of the network should be correctly displayed in a figure\n'])
+
+dcount = 0;
+for i = 1:nnodes
+    if node_sizes(i) ~= 1
+        dcount = dcount + 1;
+    end
+end
+discrete = zeros(1,dcount);
+dcount = 0;
+for i = 1:nnodes
+    if node_sizes(i) ~= 1
+        dcount = dcount + 1;
+        discrete(dcount) = i;
+    end
+end
+
+bnet = mk_bnet(dag,node_sizes,'discrete',discrete,'names',labels);
+
+%bnet.dag
+
+checkDiscreteNodes(bnet,cases);
+
+% standardize continuous data to have a mean = 0 and std = 1
+if (std_flag)
+    [cases] = standardizeData(labels,node_sizes,cases);
+end
+        
+
+end
+%  end of readInput.m
\ No newline at end of file
diff --git a/BNW_parameter_learning/readInputData.m b/BNW_parameter_learning/readInputData.m
new file mode 100644
index 00000000..2df158f9
--- /dev/null
+++ b/BNW_parameter_learning/readInputData.m
@@ -0,0 +1,75 @@
+function  [ labels , node_sizes, cases, data] = readInputData( dfile , nnodes )
+	%  readColData  reads data from a file containing data in columns
+	%               that have text titles, and possibly other header text
+	%   
+	%  Input:
+	%     dfile  = name of the file containing the data.(required)
+	%     nnodes  = number of columns in the data file.  (required)
+    %
+    %   Function assumes the following format for the input file:
+    %       1) First line has labels for each of the nodes.  There cannot
+    %              be spaces in any node label.
+    %       2) The next line is the "node_sizes" of the nodes.  If the 
+    %           nodes are discrete, this number will be equal to the number
+    %           of states.  If the nodes are continuous, they should be 
+    %           equal to 1.  The function assumes that any nodes with
+    %           node_size = 1 is continuous.
+    %       3) The rest of the file is numeric data.  The data in the input
+    %               data has the number of columns equal to the number of 
+    %               nodes in the network and the number of rows equal to
+    %               the number of samples.
+    %
+    %
+	%  Output:
+	%     labels = cell array with node (column) labels.
+    %     node_sizes  =  vector with the size of each node
+    %     cases = cell array with the data.  The cases array is transposed
+    %       in comparison with the input data to agree with the format of
+    %       cell data used in BNT.
+
+%  open file for input, include error handling
+fin = fopen(dfile,'r');
+if fin < 0
+   error(['Could not open ',dfile,' for input']);
+end
+
+% Read in first line to get the node labels.
+labels = cell(1,nnodes);
+buffer = fgetl(fin);    %get header line as a string
+for j=1:nnodes
+    [next,buffer] = strtok(buffer);
+    labels{j} = next;
+end
+
+%  Read in the data.  Use the vetorized fscanf function to load all
+%  numerical values into one vector.  Then reshape this vector into a
+%  matrix.
+
+data = fscanf(fin,'%f');  %  Load the numerical values into one long vector
+
+
+
+
+nd = length(data);        %  total number of data points
+nr = nd/nnodes;            %  number of rows; check (next statement) to make sure
+if nr ~= round(nd/nnodes)
+   fprintf(1,'\ndata: nrow = %f\tncol = %d\n',nr,nnodes);
+   fprintf(1,'number of data points = %d does not equal nrow*ncol\n',nd);
+   error('data is not rectangular')
+end
+
+data = reshape(data,nnodes,nr)';   %  have to transpose the reshaped array
+
+
+node_sizes = zeros(1,nnodes);
+for j = 1:nnodes
+    node_sizes(j) = data(1,j);
+end
+
+nr = nr - 1;
+data(1,:) = [];
+cases = cell(nnodes,nr);
+cases(:,:) = num2cell(data');
+
+end
+%  end of readInputData.m
\ No newline at end of file
diff --git a/BNW_parameter_learning/readInputStructure.m b/BNW_parameter_learning/readInputStructure.m
new file mode 100644
index 00000000..72c39f46
--- /dev/null
+++ b/BNW_parameter_learning/readInputStructure.m
@@ -0,0 +1,72 @@
+function [ dag ] = readInputStructure( sfile, labels )
+%readInputStructure Read in file with structure information
+    %   
+    %Input:
+	%     sfile  = name of the file containing the data (required)
+	%     labels = cell array with node labels. (required)
+    %     nnodes  = number of columns in the data file. (required)  
+    %
+    %   Function assumes the following format for the structure input file:
+    %       1) The first line has node labels.  These must be the same as 
+    %           in the input data file.  They cannot contain spaces.
+    %       2) The remainder of the file contains the structure of the dag.
+    %           The structure of a graph is a N-by-N matrix, where N is the
+    %           number of nodes.  There are 1's in the matrix representing
+    %           parent-child relationships.  For each 1, the row indicates
+    %           the parent and the column indicates the child.  For
+    %           example, a 1 in the (2,3) position of the matrix indicates
+    %           that there is an arc pointing from node 2 to node 3.
+    %        
+    %
+	%  Output:
+    %     dag = matrix with the structure.
+%
+%   Read in first line of the structure file
+%  open file for input, include error handling
+fin = fopen(sfile,'r');
+if fin < 0
+   error(['Could not open ',sfile,' for input']);
+end
+
+nnodes = size(labels,2);
+% Read in first line to get the node labels.
+labels_test = cell(1,nnodes);
+buffer = fgetl(fin);    %get header line as a string
+for j=1:nnodes
+    [next,buffer] = strtok(buffer);
+    labels_test{j} = next;  
+end
+    
+for j=1:nnodes
+    if labels_test{j} ~= labels{j}
+        fprintf(['Label of node ',j,' is not consistent in input and structure files'])
+    end
+end
+
+data = fscanf(fin,'%f');
+  
+nd = length(data);        %  total number of data points
+nr = nd/nnodes;            %  number of rows; check (next statement) to make sure
+if nr ~= round(nd/nnodes)
+   fprintf(1,'\ndata: nrow = %f\tncol = %d\n',nr,nnodes);
+   fprintf(1,'number of data points = %d does not equal nrow*ncol\n',nd);
+   error('Structure file does not have the correct dimensions (1)')
+end
+% check to make sure that structure is square
+if nr ~= nnodes
+    error('Structure file does not have the correct dimensions (2)')
+end
+
+data = reshape(data,nnodes,nr)';   %  have to transpose the reshaped array
+
+
+dag = zeros(nnodes,nnodes);
+for i = 1:size(data,1)
+    for j = 1:size(data,2)
+        dag(i,j) = data(i,j);
+    end
+end
+
+
+end
+%  end of readInputStructure.m
diff --git a/BNW_parameter_learning/runBN_initial.m b/BNW_parameter_learning/runBN_initial.m
new file mode 100644
index 00000000..c3f2a34b
--- /dev/null
+++ b/BNW_parameter_learning/runBN_initial.m
@@ -0,0 +1,98 @@
+function runBN_initial(pre)
+sfile=strcat(pre,'structure_input.txt');
+dfile=strcat(pre,'continuous_input.txt');
+
+nnodefile=strcat(pre,'nnode.txt');
+fnnode = fopen(nnodefile,'r');
+nnodes = fscanf(fnnode,'%d');
+
+
+mapfilename=strcat(pre,'mapdata.txt');
+mapvalfilename=strcat(pre,'map.txt');
+
+mapfile = fopen(mapfilename,'w');
+
+mapval = fopen(mapvalfilename,'w');
+
+
+%nnodes=5;
+Std_flag=true;
+[labels,cases,bnet,node_sizes,data,labelsold]=readInput(dfile,sfile,nnodes,Std_flag);
+s = std(data,0,1);
+m=mean(data);
+
+
+for i=1:nnodes
+  fprintf(mapval,'%s\t%d\t%f\t%f\n',labelsold{i},node_sizes(i),s(i),m(i));
+end
+
+
+% for j=1:nnodes
+%     [next,buffer] = strtok(buffer);
+%     name{j}=next;
+%     for i=1:nnodes    
+%         if strcmp(name{j},labels{i})
+%             map{j}=i;
+%             fprintf(mapfile,'%d\t',i);
+%         end
+%      end
+% end
+%name
+%labels
+%map
+fprintf(mapfile,'%s',labels{1});
+for i=2:nnodes
+  fprintf(mapfile,'\t%s',labels{i});
+end
+fprintf(mapfile,'\n');
+
+[bnet]=parameterLearning(bnet,cases);
+%[predict_mean,predict_sd,q_sq]=looCrossValid(bnet,cases);
+%engine=jtree_inf_engine(bnet);
+%evidence=cell(1,nnodes);
+
+%varfile='var.txt';
+%fvar = fopen(varfile,'r');
+%select_var = fscanf(fvar,'%d');
+%select_var=map{select_var};
+%varfiled='vardata.txt';
+%fvard = fopen(varfiled,'r');
+%select_var_data = fscanf(fvard,'%f');
+
+%evidence{select_var}=select_var_data;
+%[engine,loglik]=enter_evidence(engine,evidence);
+
+%outdata='prediction.txt';
+%fout = fopen(outdata,'w');
+
+%for ii = 1:nnodes 
+ % i=map{ii};
+ % data=marginal_nodes(engine,i);
+ % fprintf(fout,'%d\t%d\t%f\t%f\t%f\n',ii,data.domain,data.T,data.mu,data.Sigma);
+  %fprintf(1,'%d\n',i);
+% end
+
+filename=strcat(pre,'net_figure.txt');
+drawFigure(nnodes,bnet,labels,filename,cases);
+
+%quit force;
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{1}=2;
+%[engine,loglik]=enter_evidence(engine,evidence)
+%marginal_nodes(engine,1)
+%marginal_nodes(engine,2)
+%marginal_nodes(engine,3)
+%marginal_nodes(engine,4)
+%marginal_nodes(engine,5)
+%evidence{2}=0.6;
+%evidence{1}=[];
+%[engine,loglik]=enter_evidence(engine,evidence);
+%marginal_nodes(engine,3);
+%marginal_nodes(engine,4);
+%marginal_nodes(engine,5);
+fclose(mapval);
+fclose(mapfile);
+end
\ No newline at end of file
diff --git a/BNW_parameter_learning/standardizeData.m b/BNW_parameter_learning/standardizeData.m
new file mode 100644
index 00000000..ba4ef5dd
--- /dev/null
+++ b/BNW_parameter_learning/standardizeData.m
@@ -0,0 +1,25 @@
+function [ cases ] = standardizeData( labels, node_sizes, cases )
+%standardizeData standardizes continuous nodes so they have a mean = 0
+%   and standard deviation = 1
+%   Detailed explanation goes here
+
+nnodes = size(labels,2);
+
+%fprintf(['Standardizing data for continuous nodes\n'])
+for i = 1:nnodes
+    if node_sizes(i) == 1
+        temp = cell2num(cases(i,:));
+        [temp] = standardize(temp);
+        cases(i,:) = num2cell(temp);
+    end
+end
+
+%write standardized data to file
+%fprintf(['Standardized data is written to file standardized_data.txt\n'])
+%fout = 'standardized_data.txt';
+%txt = sprintf([repmat('%s\t',1,size(labels,2))],labels{:});
+%dlmwrite(fout,txt,'');
+%dlmwrite(fout,cell2num(cases'),'-append','delimiter','\t');
+
+end
+
author	ziejd2	2017-09-14 16:17:47 -0500
committer	GitHub	2017-09-14 16:17:47 -0500
commit	7cc31810d53176e805532b2789955f4eedbce6bb (patch)
tree	82924642070d871f753ee41c0f3e363ff8f380da
parent	6882395afdadf4e982b25b5215071a0932730950 (diff)
download	BNW-7cc31810d53176e805532b2789955f4eedbce6bb.tar.gz