Skip to content

Commit

Permalink
Add files via upload
Browse files Browse the repository at this point in the history
  • Loading branch information
@JingweiToo
JingweiToo authored Dec 19, 2020
1 parent 30993f3 commit 17f068a
Show file tree
Hide file tree
Showing 3 changed files with 155 additions and 123 deletions.
51 changes: 26 additions & 25 deletions Main.m
View file
Original file line number Diff line number Diff line change
@@ -1,42 +1,43 @@
%-------------------------------------------------------------------------%
% Binary Harris Hawk Optimization (BHHO) source codes demo version %
% %
% Programmer: Jingwei Too %
% %
% E-Mail: jamesjames868@gmail.com %
%-------------------------------------------------------------------------%
%-------------------------------------------------------------------%
% Binary Harris Hawk Optimization (BHHO) demo version %
%-------------------------------------------------------------------%


%---Inputs-----------------------------------------------------------------
% feat: features
% label: labelling
% N: Number of hawks
% T: Maximum number of iterations
%---Outputs----------------------------------------------------------------
% sFeat: Selected features
% Sf: Selected feature index
% Nf: Number of selected features
% curve: Convergence curve
%--------------------------------------------------------------------------
%---Inputs-----------------------------------------------------------
% feat : feature vector (instances x features)
% label : label vector (instance x 1)
% N : Number of hawks
% max_Iter : Maximum number of iterations

%---Outputs----------------------------------------------------------
% sFeat : Selected features
% Sf : Selected feature index
% Nf : Number of selected features
% curve : Convergence curve
%--------------------------------------------------------------------


%% Binary Harris Hawk Optimization
clc, clear, close;
% Benchmark data set
load ionosphere.mat;

% Set 20% data as validation set
ho=0.2;
ho = 0.2;
% Hold-out method
HO=cvpartition(label,'HoldOut',ho,'Stratify',false);
HO = cvpartition(label,'HoldOut',ho,'Stratify',false);

% Parameter setting
N=10; T=100;
N = 10;
max_Iter = 100;
% Binary Harris Hawk Optimization
[sFeat,Sf,Nf,curve]=jBHHO(feat,label,N,T,HO);
% Plot convergence curve
figure(); plot(1:T,curve); xlabel('Number of iterations');
ylabel('Fitness Value'); title('BHHO'); grid on;
[sFeat,Sf,Nf,curve] = jBHHO(feat,label,N,max_Iter,HO);

% Plot convergence curve
plot(1:max_Iter,curve);
xlabel('Number of iterations');
ylabel('Fitness Value');
title('BHHO'); grid on;



192 changes: 109 additions & 83 deletions jBHHO.m
View file
Original file line number Diff line number Diff line change
@@ -1,144 +1,170 @@
function [sFeat,Sf,Nf,curve]=jBHHO(feat,label,N,T,HO)
function [sFeat,Sf,Nf,curve] = jBHHO(feat,label,N,max_Iter,HO)

fun=@jFitnessFunction;
D=size(feat,2); X=zeros(N,D);
for i=1:N
for d=1:D
beta = 1.5;
ub = 1;
lb = 0;

fun = @jFitnessFunction;
dim = size(feat,2);
X = zeros(N,dim);
for i = 1:N
for d = 1:dim
if rand() > 0.5
X(i,d)=1;
X(i,d) = 1;
end
end
end
fitR=inf; fit=zeros(1,N); Y=zeros(1,D); Z=zeros(1,D);
beta=1.5; ub=1; lb=0; t=1; curve=inf;
%---Iteration start-------------------------------------------------------
while t <= T
for i=1:N
fit(i)=fun(feat,label,X(i,:),HO);
fitR = inf;
fit = zeros(1,N);
Y = zeros(1,dim);
Z = zeros(1,dim);

curve = inf;
t = 1;
%---Iteration start-------------------------------------------------
while t <= max_Iter
for i = 1:N
fit(i) = fun(feat,label,X(i,:),HO);
if fit(i) < fitR
fitR=fit(i); Xrb=X(i,:);
fitR = fit(i);
Xrb = X(i,:);
end
end
Xm=mean(X,1);
for i=1:N
E0=-1+2*rand();
E=2*E0*(1-(t/T));
X_mu = mean(X,1);
for i = 1:N
E0 = -1 + 2 * rand();
E = 2 * E0 * (1 - (t / max_Iter));
if abs(E) >= 1
q=rand();
q = rand();
if q >= 0.5
k=randi([1,N]); r1=rand(); r2=rand();
for d=1:D
Xn=X(k,d)-r1*abs(X(k,d)-2*r2*X(i,d));
S=1/(1+exp(-Xn));
k = randi([1,N]);
r1 = rand();
r2 = rand();
for d = 1:dim
Xn = X(k,d) - r1 * abs(X(k,d) - 2 * r2 * X(i,d));
S = 1 / (1 + exp(-Xn));
if rand() < S
X(i,d)=1;
else
X(i,d)=0;
X(i,d)= 1;
else
X(i,d) = 0;
end
end
elseif q < 0.5
r3=rand(); r4=rand();
for d=1:D
Xn=(Xrb(d)-Xm(d))-r3*(lb+r4*(ub-lb));
S=1/(1+exp(-Xn));
r3 = rand();
r4 = rand();
for d = 1:dim
Xn = (Xrb(d) - X_mu(d)) - r3 * (lb + r4 * (ub - lb));
S = 1 / (1 + exp(-Xn));
if rand() < S
X(i,d)=1;
X(i,d) = 1;
else
X(i,d)=0;
X(i,d) = 0;
end
end
end
elseif abs(E) < 1
J=2*(1-rand()); r=rand();
if r >= 0.5 && abs(E) >= 0.5
for d=1:D
DX=Xrb(d)-X(i,d);
Xn=DX-E*abs(J*Xrb(d)-X(i,d));
S=1/(1+exp(-Xn));
J = 2 * (1 - rand());
r = rand();
if r >= 0.5 && abs(E) >= 0.5
for d = 1:dim
DX = Xrb(d) - X(i,d);
Xn = DX - E * abs(J * Xrb(d) - X(i,d));
S = 1 / (1 + exp(-Xn));
if rand() < S
X(i,d)=1;
X(i,d) = 1;
else
X(i,d)=0;
X(i,d) = 0;
end
end
elseif r >= 0.5 && abs(E) < 0.5
for d=1:D
DX=Xrb(d)-X(i,d);
Xn=Xrb(d)-E*abs(DX);
S=1/(1+exp(-Xn));
elseif r >= 0.5 && abs(E) < 0.5
for d = 1:dim
DX = Xrb(d) - X(i,d);
Xn = Xrb(d) - E * abs(DX);
S = 1 / (1 + exp(-Xn));
if rand() < S
X(i,d)=1;
X(i,d) = 1;
else
X(i,d)=0;
X(i,d) = 0;
end
end
elseif r < 0.5 && abs(E) >= 0.5
LF=jLevyDistribution(beta,D);
for d=1:D
Yn=Xrb(d)-E*abs(J*Xrb(d)-X(i,d));
S=1/(1+exp(-Yn));
elseif r < 0.5 && abs(E) >= 0.5
LF = jLevyDistribution(beta,dim);
for d = 1:dim
Yn = Xrb(d) - E * abs(J * Xrb(d) - X(i,d));
S = 1 / (1 + exp(-Yn));
if rand() < S
Y(d)=1;
Y(d) = 1;
else
Y(d)=0;
Y(d) = 0;
end
Zn=Y(d)+rand()*LF(d);
S=1/(1+exp(-Zn));
Zn = Y(d) + rand() * LF(d);
S = 1 / (1 + exp(-Zn));
if rand() < S
Z(d)=1;
Z(d) = 1;
else
Z(d)=0;
Z(d) = 0;
end
end
fitY=fun(feat,label,Y,HO); fitZ=fun(feat,label,Z,HO);
fitY = fun(feat,label,Y,HO);
fitZ = fun(feat,label,Z,HO);
if fitY <= fit(i)
fit(i)=fitY; X(i,:)=Y;
fit(i) = fitY;
X(i,:) = Y;
end
if fitZ <= fit(i)
fit(i)=fitZ; X(i,:)=Z;
fit(i) = fitZ;
X(i,:) = Z;
end
elseif r < 0.5 && abs(E) < 0.5
LF=jLevyDistribution(beta,D);
for d=1:D
Yn=Xrb(d)-E*abs(J*Xrb(d)-Xm(d));
S=1/(1+exp(-Yn));
elseif r < 0.5 && abs(E) < 0.5
LF = jLevyDistribution(beta,dim);
for d = 1:dim
Yn = Xrb(d) - E * abs(J * Xrb(d) - X_mu(d));
S = 1 / (1 + exp(-Yn));
if rand() < S
Y(d)=1;
Y(d) = 1;
else
Y(d)=0;
Y(d) = 0;
end
Zn=Y(d)+rand()*LF(d);
S=1/(1+exp(-Zn));
Zn = Y(d) + rand() * LF(d);
S = 1 / (1 + exp(-Zn));
if rand() < S
Z(d)=1;
Z(d) = 1;
else
Z(d)=0;
Z(d) = 0;
end
end
fitY=fun(feat,label,Y,HO); fitZ=fun(feat,label,Z,HO);
fitY = fun(feat,label,Y,HO);
fitZ = fun(feat,label,Z,HO);
if fitY <= fit(i)
fit(i)=fitY; X(i,:)=Y;
fit(i) = fitY;
X(i,:) = Y;
end
if fitZ <= fit(i)
fit(i)=fitZ; X(i,:)=Z;
fit(i) = fitZ;
X(i,:) = Z;
end
end
end
end
curve(t)=fitR;
curve(t) = fitR;
fprintf('\nIteration %d Best (BHHO)= %f',t,curve(t))
t=t+1;
t = t + 1;
end
Pos=1:D; Sf=Pos(Xrb==1); Nf=length(Sf); sFeat=feat(:,Sf);
Pos = 1:dim;
Sf = Pos(Xrb == 1);
Nf = length(Sf);
sFeat = feat(:,Sf);
end


function LF=jLevyDistribution(beta,D)
nume=gamma(1+beta)*sin(pi*beta/2);
deno=gamma((1+beta)/2)*beta*2^((beta-1)/2);
sigma=(nume/deno)^(1/beta);
u=randn(1,D)*sigma; v=randn(1,D);
step=u./abs(v).^(1/beta); LF=0.01*step;
function LF = jLevyDistribution(beta,dim)
nume = gamma(1 + beta) * sin(pi * beta / 2);
deno = gamma((1 + beta) / 2) * beta * 2 ^ ((beta - 1) / 2);
sigma = (nume / deno) ^ (1 / beta);
u = randn(1,dim) * sigma;
v = randn(1,dim);
step = u ./ abs(v) .^ (1 / beta);
LF = 0.01 * step;
end


35 changes: 20 additions & 15 deletions jFitnessFunction.m
View file
Original file line number Diff line number Diff line change
@@ -1,28 +1,33 @@
% Notation: This fitness function is for demonstration

function fitness=jFitnessFunction(feat,label,X,HO)
if sum(X==1)==0
fitness=inf;
function cost = jFitnessFunction(feat,label,X,HO)
if sum(X == 1) == 0
cost = inf;
else
fitness=jwrapperKNN(feat(:,X==1),label,HO);
cost = jwrapperKNN(feat(:, X == 1),label,HO);
end
end


function ER=jwrapperKNN(sFeat,label,HO)
function error = jwrapperKNN(sFeat,label,HO)
%---// Parameter setting for k-value of KNN //
k=5;
xtrain=sFeat(HO.training==1,:); ytrain=label(HO.training==1);
xvalid=sFeat(HO.test==1,:); yvalid=label(HO.test==1);
Model=fitcknn(xtrain,ytrain,'NumNeighbors',k);
pred=predict(Model,xvalid);
N=length(yvalid); correct=0;
for i=1:N
k = 5;

xtrain = sFeat(HO.training == 1,:);
ytrain = label(HO.training == 1);
xvalid = sFeat(HO.test == 1,:);
yvalid = label(HO.test == 1);

Model = fitcknn(xtrain,ytrain,'NumNeighbors',k);
pred = predict(Model,xvalid);
num_valid = length(yvalid);
correct = 0;
for i = 1:num_valid
if isequal(yvalid(i),pred(i))
correct=correct+1;
correct = correct + 1;
end
end
Acc=correct/N;
ER=1-Acc;
Acc = correct / num_valid;
error = 1 - Acc;
end

0 comments on commit 17f068a

Please sign in to comment.