function [R,P,area]=PND_show_ROC1(name_of_file,feature,s)
  S=load([name_of_file,'.mat']);
  lambda_min=min(S.X(:,feature)); 
  lambda_max=max(S.X(:,feature)); 
  lambda_N=100;
  lambda_l=lambda_min+(1:lambda_N)*(lambda_max-lambda_min)/lambda_N;
  info.sign=s; info.feature=feature; 
  R=zeros(length(lambda_l),1);
  P=zeros(length(lambda_l),1);
  for lambda_=1:length(lambda_l)
    lambda=lambda_l(lambda_);
    info.lambda=lambda;
    y_hat=zeros(size(S.X,1),1);
    for n=1:size(S.X,1)
      y_hat(n)=PNC_predict1(info,S.X(n,:));
    end
    R(lambda_)=sum((y_hat==1).*(y==1))/sum(y==1);
    P(lambda_)=sum((y_hat==1).*(y==1))/sum(y_hat==1);
  end
  area=sum((1-P(2:end)-1+P(1:(end-1)).*(R(1:end-1)+R(2:end))/2);
end

figure(1); plot(1-P,R);

feature
lambda
sign

x*sign<lambda*sign


  lambda_min=min(X); 
  lambda_max=max(X); 
  lambda_N=100;
  lambda_l=lambda_min+(1:lambda_N)*(lambda_max-lambda_min)/lambda_N;
  s=1; 
  N=500;
  R=zeros(length(lambda_l),1);
  P=zeros(length(lambda_l),1);
  for lambda_=1:length(lambda_l)
    lambda=lambda_l(lambda_);
    info.lambda=lambda;
    y_hat=zeros(N,1);
    for n=1:N
      y_hat(n)=(X(n)>=lambda);
    end
    R(lambda_)=sum((y_hat==1).*(Y==1))/sum(Y==1);
    P(lambda_)=sum((y_hat==1).*(Y==1))/sum(y_hat==1);
  end
  area
end



function info=PND_train1(feature_l)
  info.intercept=-1;
  S=load('training.mat'); 
  info.weights=zeros(1,S.metadata.F)
  for feature_=1:length(feature_l)
    f=feature_l(feature_);
    w_max=1/min(S.X(:,f));
    w_min=1/max(S.X(:,f));
    w=w_min+rand(1)*(w_max-w_min);
    info.weights(f)=w;
  end
end

function y_hat=PND_predict1(info,x)
  y_hat=(sum(info.weight.*x)+info.intercept>0);
end


function info=PND_train2(feature_l)
  exp_n=200;
  info=PND_train1(feature_l);
  score=PNC_score1(info,'training',@PND_predict1);
  OA=score.overall_accuracy;
  for exp=1:exp_n
    info_try=PND_train1(feature_l);
    score=PNC_score1(info_try,'training',@PND_predict1);
    OA_try=score.overall_accuracy;
    if OA_try>OA
      info=info_try; 
      OA=OA_try; 
      disp(['OA=',num2str(OA),' exp=',num2str(exp_n)]),
    end
  end
end

binary_dataset=PNC_BinaryDataset(4);
score=PNC_score2(binary_dataset,@PND_train2,@PND_predict1,5);


function info=PND_train3()
  S=load('training');
  Xe=[S.X ones(size(S.X,1),1)];
  Xe_inv=pinv(Xe); %inv(Xe'*Xe)*Xe';
  Ae=Xe_inv*S.Y; 
  info.weights=reshape(Ae(1:end-1),1,S.metadata.F); info.intercep=Ae(end);
end