浏览 2594 次
|
锁定老帖子 主题：神经网络和遗传算法（NN&GA）
精华帖 (0) :: 良好帖 (7) :: 新手帖 (0) :: 隐藏帖 (0)
|
|
|---|---|
| 作者 | 正文 |
|
发表时间：2008-04-20
在软计算课程上，老师让我们编程实现Hopfield网、联想记忆、BM、BP、SOFM、遗传算法。正在学习Ruby，就用ruby实现了上述内容。上述实现的内容比较简单，但没有对每个程序的输入和输出做说明。如果有任何疑问可以留言。希望和大家交流！
Hopfield w = Array.new(4)
4.times do |i|
w[i] = Array.new(4,0)
4.times do |j|
if j > i
w[i][j] = (rand 101) / 100.0 * (-1) ** (rand 2)
else
w[i][j]=w[j][i]
end
end
end
u=Array.new(4)
4.times do |i|
u[i] = (rand 101) / 100.0 * (-1) ** (rand 2)
end
target = Array.new(16)
16.times do |i|
target[i] = Array.new(4)
if i == 0
target[0] = [-1,-1,-1,-1]
next
end
if i > 0
4.times do |k|
t = target[i-1][k]
target[i][k] = t
end
4.times do |j|
ttt = target[i-1][3-j]
if ttt == -1
target[i][3-j] = 1
break
else
target[i][3-j] = -1
end
end
end
end
16.times do |k|
energy = 0
4.times do |i|
4.times do |j|
energy += -0.5*w[i][j]*target[k][i]*target[k][j]
end
energy+=u[i]*target[k][i]
end
puts target[k].join(" ") +"\t\t"+ energy.to_s
end
16.times do |i|
tem = ""
puts "\n"
4.times do |k|
tem += target[i][k].to_s + " "
end
puts tem
tem = ""
total = 0
30.times do |j|
total = 0
4.times do |k|
kk = target[i][k]
if kk == 1
total += 2 ** (3-k)
end
end
if total/10 == 1
tem += total.to_s + " "
else
tem += " " + total.to_s + " "
end
total=0
rRound = rand 4
4.times do |k|
total += w[rRound][k] * target[i][k]
end
if total-u[rRound] >= 0
target[i][rRound] = 1
else
target[i][rRound] = -1
end
end
total = 0
4.times do |j|
4.times do |k|
total += -0.5 * w[j][k] * target[i][j] * target[i][k]
end
total += u[j] * target[i][j]
end
tem += " " + total.to_s
puts tem
end
BM算法 initial=[1,-1,1,-1]
t,n=1,1000
w = Array.new(4)
4.times do |i|
w[i] = Array.new(4,0)
4.times do |j|
if j > i
w[i][j] = (rand 101) / 100.0 * (-1) ** (rand 2)
else
w[i][j]=w[j][i]
end
end
end
u=Array.new(4)
4.times do |i|
u[i] = (rand 101) / 100.0 * (-1) ** (rand 2)
end
def p(tt,mm)
1.0/(1+Math.exp( -tt/mm ))
end
while n>0.01
select=rand 4
hi=0
4.times do |j|
hi+=w[select][j]*initial[j]
end
hi-=u[select]
if hi>0
initial[select]=1
else
a=p(hi,n)
if n>10
b=(rand 700)/1000.0
else
b=0.5+(rand 200)/1000.0
end
if a > b
initial[select]=1
else
initial[select]=-1
end
end
n=1000/(1+t)
t+=1
end
total=0
4.times do |i|
4.times do |j|
total+=-0.5*w[i][j]*initial[i]*initial[j]
end
total+=u[i]*initial[i]
end
puts initial.join("\t")+"\t\t"+total.to_s
puts "\n"
target = Array.new(16)
16.times do |i|
target[i] = Array.new(4)
if i == 0
target[0] = [-1,-1,-1,-1]
next
end
if i > 0
4.times do |k|
t = target[i-1][k]
target[i][k] = t
end
4.times do |j|
ttt = target[i-1][3-j]
if ttt == -1
target[i][3-j] = 1
break
else
target[i][3-j] = -1
end
end
end
end
16.times do |k|
energy = 0
4.times do |i|
4.times do |j|
energy += -0.5*w[i][j]*target[k][i]*target[k][j]
end
energy+=u[i]*target[k][i]
end
puts target[k].join(" ") +"\t\t"+ energy.to_s
end
16.times do |i|
tem = ""
puts "\n"
4.times do |k|
tem += target[i][k].to_s + " "
end
puts tem
tem = ""
total = 0
30.times do |j|
total = 0
4.times do |k|
kk = target[i][k]
if kk == 1
total += 2 ** (3-k)
end
end
if total/10 == 1
tem += total.to_s + " "
else
tem += " " + total.to_s + " "
end
total=0
rRound = rand 4
4.times do |k|
total += w[rRound][k] * target[i][k]
end
if total-u[rRound] >= 0
target[i][rRound] = 1
else
target[i][rRound] = -1
end
end
total = 0
4.times do |j|
4.times do |k|
total += -0.5 * w[j][k] * target[i][j] * target[i][k]
end
total += u[j] * target[i][j]
end
tem += " " + total.to_s
puts tem
end
BP算法 #BP Algorithm endeavor
def f(t)
m=1.0/(1.0+Math.exp(-t))
end
num_mode=8
num_input_nerve=3
num_middle_nerve=20
num_output_nerve=2
num_learntimes=200
affix=0.5
d=Array.new(num_mode)
num_mode.times do |i|
d[i]=Array.new(num_output_nerve)
num_output_nerve.times do |j|
d[i][j]=(rand 101)/100.0
end
end
input=Array.new(num_mode)
num_mode.times do |i|
input[i]=Array.new(num_input_nerve)
num_input_nerve.times do |j|
input[i][j]=(rand 101)/100.0
end
end
v=Array.new(num_input_nerve)
num_input_nerve.times do |i|
v[i]=Array.new(num_middle_nerve)
num_middle_nerve.times do |j|
v[i][j]=(rand 101)/100.0*(-1)**(rand 2)
end
end
r=Array.new(num_middle_nerve)
num_middle_nerve.times do |i|
r[i]=(rand 101)/100.0*(-1)**(rand 2)
end
middle=Array.new()
w=Array.new(num_middle_nerve)
num_middle_nerve.times do |i|
w[i]=Array.new(num_output_nerve)
num_output_nerve.times do |j|
w[i][j]=(rand 101)/100.0*(-1)**(rand 2)
end
end
s=Array.new(num_middle_nerve)
num_middle_nerve.times do |i|
s[i]=(rand 101)/100.0*(-1)**(rand 2)
end
output=Array.new()
#starts learning
num_learntimes.times do |g|
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
help_w=Array.new(num_output_nerve,0)
help_v=Array.new(num_middle_nerve,0)
#figure out results
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
end
#adjustment
num_output_nerve.times do |i|
help_w[i]=(d[k][i]-output[i])*output[i]*(1-output[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
help_v[i]+=help_w[j]*w[i][j]
end
help_v[i]*=middle[i]*(1-middle[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
w[i][j]+=affix*help_w[j]*middle[i]
end
end
num_output_nerve.times do |i|
s[i]-=affix*help_w[i]
end
num_input_nerve.times do |i|
num_middle_nerve.times do |j|
v[i][j]+=affix*help_v[j]*input[k][i]
end
end
num_middle_nerve.times do |i|
r[i]-=affix*help_v[i]
end
end
#end learn 1 time
end
#test on learning times
errorsum=0
puts "results:"
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
errorsum+=(d[k][i]-output[i])**2
end
puts "%.2f" %output[0]+" "+"%.2f" %output[1]
end
errorsum/=2
puts "total deviation:"+errorsum.to_s
num_learntimes=100
#starts learning
num_learntimes.times do |g|
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
help_w=Array.new(num_output_nerve,0)
help_v=Array.new(num_middle_nerve,0)
#figure out results
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
end
#adjustment
num_output_nerve.times do |i|
help_w[i]=(d[k][i]-output[i])*output[i]*(1-output[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
help_v[i]+=help_w[j]*w[i][j]
end
help_v[i]*=middle[i]*(1-middle[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
w[i][j]+=affix*help_w[j]*middle[i]
end
end
num_output_nerve.times do |i|
s[i]-=affix*help_w[i]
end
num_input_nerve.times do |i|
num_middle_nerve.times do |j|
v[i][j]+=affix*help_v[j]*input[k][i]
end
end
num_middle_nerve.times do |i|
r[i]-=affix*help_v[i]
end
end
#end learn 1 time
end
#test on learning times
errorsum=0
puts "results:"
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
errorsum+=(d[k][i]-output[i])**2
end
puts "%.2f" %output[0]+" "+"%.2f" %output[1]
end
errorsum/=2
puts "total deviation:"+errorsum.to_s
num_learntimes=100
#starts learning
num_learntimes.times do |g|
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
help_w=Array.new(num_output_nerve,0)
help_v=Array.new(num_middle_nerve,0)
#figure out results
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
end
#adjustment
num_output_nerve.times do |i|
help_w[i]=(d[k][i]-output[i])*output[i]*(1-output[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
help_v[i]+=help_w[j]*w[i][j]
end
help_v[i]*=middle[i]*(1-middle[i])
end
num_middle_nerve.times do |i|
num_output_nerve.times do |j|
w[i][j]+=affix*help_w[j]*middle[i]
end
end
num_output_nerve.times do |i|
s[i]-=affix*help_w[i]
end
num_input_nerve.times do |i|
num_middle_nerve.times do |j|
v[i][j]+=affix*help_v[j]*input[k][i]
end
end
num_middle_nerve.times do |i|
r[i]-=affix*help_v[i]
end
end
#end learn 1 time
end
#test on learning times
errorsum=0
puts "results:"
num_mode.times do |k|
middle=Array.new(num_middle_nerve,0)
output=Array.new(num_output_nerve,0)
num_middle_nerve.times do |i|
num_input_nerve.times do |j|
middle[i]+=input[k][j]*v[j][i]
end
middle[i]=f(middle[i]-r[i])
end
num_output_nerve.times do |i|
num_middle_nerve.times do |j|
output[i]+=middle[j]*w[j][i]
end
output[i]=f(output[i]-s[i])
errorsum+=(d[k][i]-output[i])**2
end
puts "%.2f" %output[0]+" "+"%.2f" %output[1]
end
puts "want to be:"
num_mode.times do |k|
puts "%.2f" %d[k][0]+" "+"%.2f" %d[k][1]
end
errorsum/=2
puts "total deviation:"+errorsum.to_s
联想记忆（Hopfield） def ppp( aa )
note = ""
15.times do |i|
if aa[i] == 1
note += " 1\t"
else
note += "-1\t"
end
end
note
end
def change( bb, cc )
15.times do |i|
if bb[i] == 1
if ((rand 2) + (rand 2)) / 2 == 0
cc[i] = 1
else
next
end
else
next
end
end
cc
end
u=Array.new(15)
15.times do |i|
u[i]=(rand 101)/100.0*(-1)**(rand 2)
end
w = Array.new(15,0)
15.times do |i|
w[i] = Array.new(15,0)
15.times do |j|
if j > i
w[i][j]=(rand 101)/100.0*(-1)**(rand 2)
else
w[i][j]=w[j][i]
end
end
end
# inital status
ini_status = Array.new(15,0)
15.times do |i|
if (rand 2) == 0
ini_status[i] = -1
else
ini_status[i] = 1
end
end
puts "Forehead status",ppp(ini_status)
120.times do |i|
now = rand 15
tt = 0
15.times do |j|
tt += w[now][j] * ini_status[j]
end
if tt-u[now] >= 0
ini_status[now] = 1
else
ini_status[now] = -1
end
end
puts "\nMemory status(->forehead status)",ppp(ini_status)
temp = 0
15.times do |i|
15.times do |j|
temp += -0.5 * w[i][j] * ini_status[i] * ini_status[j]
end
end
puts "Engergy is" + temp.to_s
target = [-1,1,-1,1,-1,1,1,1,1,1,-1,1,1,-1,1]
puts "\nA letter target(what we want to become)",ppp(target)
test = Array.new(15,-1)
test = [-1,-1,-1,1,-1,1,-1,-1,-1,1,-1,-1,-1,-1,-1]
puts "\nTest object(we want it to become A letter)",ppp(test)
simulate = Array.new(15,0)
15.times do |i|
if test[i] != target[i]
simulate[i] = -1 * ini_status[i]
else
simulate[i] = ini_status[i]
end
end
puts "\n#Hopfield new initial status(->memory status(generated by above 3))",ppp(simulate)
120.times do |i|
now = rand 15
tt = 0
15.times do |j|
tt += w[now][j] * simulate[j]
end
if tt-u[now] >= 0
simulate[now] = 1
else
simulate[now] = -1
end
end
puts "\nFinal status(->hopfield new initial status(by the same w[]))",ppp(simulate)
temp = 0
15.times do |i|
15.times do |j|
temp += -0.5 * w[i][j] * simulate[i] * simulate[j]
end
end
puts "Engergy is" + temp.to_s
judge = "Y"
15.times do |i|
if ini_status[i] != simulate[i]
judge = "N"
puts "\nMemory lost!!!"
break
end
end
if judge == "Y"
puts "\nWe learnt it!! "
end
SOFM网（输入为sin曲线。实现了点的文件生成，再用C语言画图，ruby是在不容易实现画图，除非在Mac系统里） #SOM another try!
#by Brinado on Apr. 17th, 2008
#initialization
num_mode=80
num_matrix=10
num_learn=60
ita=0.5
sigma=1
#input -- Math.sin
input=Array.new(num_mode)
num_mode.times do |i|
input[i]=Array.new(2)
2.times do |j|
input[i][0]=1.0*i/num_mode
input[i][1]=Math.sin(Math::PI*2.0*i/num_mode)/2+0.5
end
end
#w
w=Array.new(100)
100.times do |i|
w[i]=Array.new(2)
2.times do |j|
w[i][j]=0.5+(rand 100)/1000*(-1)**(rand 2)
end
end
#output
output=Array.new(num_mode)
num_mode.times do |i|
output[i]=Array.new(2)
2.times do |j|
output[i][j]=1.0*i/num_mode
output[i][j]=Math.sin(Math::PI*2.0*i/num_mode)
end
end
num_learn.times do |g|
num_mode.times do |k|
distance=Array.new(100)
a,b,tt3=0,0,999
100.times do |i|
tt=Math.sqrt((input[k][0]-w[i][0])**2+(input[k][1]-w[i][1])**2)
#puts tt
distance[i]=tt
if tt<tt3
a,tt3=i,tt
end
end
100.times do |i|
2.times do |j|
tta1,tta2=(a/9-i/9)**2+(a%9-i%9)**2,sigma**2
w[i][j]+=ita*Math.exp(-1.0*tta1/sigma)*(input[k][j]-w[i][j])
end
end
#insertion here...
end
if sigma<0.005
sigma==0.005
else
sigma-=0.005
end
end
txt = ""
f = File.open("db.txt", "w")
100.times do |i|
txt = "%.15f" %w[i][0] + " " + "%.15f" %w[i][1] + " "
f.write(txt)
end
f.close
SOFM网C语言画图实现 #include<stdio.h>
#include<stdlib.h>
#include<graphics.h>
#define MM 19
main()
{
char ch[MM];
int i=0,j=0;
int object[200];
int graphdriver=VGA;
int graphmode=VGAHI;
FILE *f;
if((f=fopen("[b]c:/turboc2/brinado/db2.txt[/b]","r"))==NULL)
{
printf("db.txt open error.\n");
exit(0);
}
while(i<200)
{
fgets(ch,MM,f);
object[i]=(int)((ch[2]-'0')*10+ch[3]-'0'+(ch[4]-'0')*0.1)*4;
i++;
}
fclose(f);
initgraph(&graphdriver,&graphmode," ");
cleardevice();
/*
setcolor(RED);
for(j=0;j<200;j+=2)
{
if(j==18||j==38||j==58||j==78||j==98||j==118||j==138||j==158||j==178||j==198)
{
if(j!=198){line(object[j],object[j+1],object[j+20],object[j+21]);}
continue;
}
line(object[j],object[j+1],object[j+2],object[j+3]);
if(j<180)
{
line(object[j],object[j+1],object[j+20],object[j+21]);
}
}
*/
for(j=0;j<200;j+=2)
{
putpixel(object[j],object[j+1],2);
putpixel(object[j]+1,object[j+1],1);
putpixel(object[j],object[j+1]+1,1);
putpixel(object[j],object[j+1]-1,1);
putpixel(object[j]-1,object[j+1],1);
}
setcolor(WHITE);
rectangle(0,0,639,479);
getch();
closegraph();
}
GA-遗传算法 #target:[0 0 0 0 0] [1 1 1 1 1]
#evaluate fitness
def godbenchmark(god)
result=0
10.times do |i|
if i<5
if god[i]==0
result+=1
end
else
if god[i]==1
result+=1
end
end
end
result #return fitness value
end
#crossover using single point method
#change from 0~10 points totally. When set position===10, operation equals to completely change 2 chrommosomes
def crossover(a,b,position)
10.times do |j|
if j>=position
tt=a[j]
a[j]=b[j]
b[j]=tt
end
end
end
#initialize chromosome
chromosome=Array.new(6)
6.times do |i|
chromosome[i]=Array.new(10,0)
10.times do |j|
if (rand 2)==1
chromosome[i][j]=1
end
end
end
#parameters
#p_crossover=0.4
#p_mutation=3/60
#num_chromosome=6
#num_gene=10
iterates=50
step=0
record=Array.new(50)
init=Array.new(6,0)
6.times do |i|
init[i]=Array.new(2)
init[i][0]=i
init[i][1]=godbenchmark(chromosome[i])
end
#pop ranking
ttt,t1=init[0][1],0
6.times do |i|
for j in 0..(5-i)
if j==0
ttt,t1=init[0][1],0
end
if init[j][1] > ttt
mm1,mm2=init[j][0],init[j][1]
init[j][0],init[j][1]=init[t1][0],init[t1][1]
init[t1][0],init[t1][1]=mm1,mm2
ttt,t1=init[j][1],j
else
ttt,t1=init[j][1],j
end
end
end
#select using tournament method
middle=Array.new(6)
6.times do |i|
middle[i]=Array.new(10)
if i==0
10.times do |j|
middle[i][j]=chromosome[init[i][0]][j]
end
else
10.times do |j|
middle[i][j]=chromosome[init[i-1][0]][j]
end
end
end
chromosome=middle
#iteration begins...
iterates.times do |k|
#crossover begin
3.times do |i|
if (rand 10)/10.0 >= 0.6
case i
when 0
crossover chromosome[0],chromosome[5],(rand 10) #must change anyway!!!
when 1
crossover chromosome[1],chromosome[4],(rand 10)
when 2
crossover chromosome[2],chromosome[3],(rand 10)
end
end
end
#crossover end
#mutation begin
6.times do |i|
percentage=0.85
if (rand 100)/100.0 >= percentage
tt=rand 10
if chromosome[i][tt]==0
chromosome[i][tt]=1
else
chromosome[i][tt]=0
end
end
end
#mutation end
#selection begin
init=Array.new(6,0)
6.times do |i|
init[i]=Array.new(2)
init[i][0]=i
init[i][1]=godbenchmark(chromosome[i])
end
#pop ranking
ttt,t1=init[0][1],0
6.times do |i|
for j in 0..(5-i)
if j==0
ttt,t1=init[0][1],0
end
if init[j][1] > ttt
mm1,mm2=init[j][0],init[j][1]
init[j][0],init[j][1]=init[t1][0],init[t1][1]
init[t1][0],init[t1][1]=mm1,mm2
ttt,t1=init[j][1],j
else
ttt,t1=init[j][1],j
end
end
end
#select using tournament method
middle=Array.new(6)
6.times do |i|
middle[i]=Array.new(10)
if i==0
10.times do |j|
middle[i][j]=chromosome[init[i][0]][j]
end
else
10.times do |j|
middle[i][j]=chromosome[init[i-1][0]][j]
end
end
end
chromosome=middle
#select end
record[step]=godbenchmark(chromosome[0])
step+=1
#exit condition
if godbenchmark(chromosome[0])==10
break
end
end
#print initial chromosome
print="--Final Result--\n"
6.times do |i|
10.times do |j|
print+=chromosome[i][j].to_s+"\t"
end
print+="\n"
end
puts print
puts "\n*Note:this result is the No.#{step} generation's reproduction!","\n*Step change(fitness result) in detail:\n","\n"
step.times do |i|
puts "No.#{i} step best: "+record[i].to_s
end
声明：ITeye文章版权属于作者，受法律保护。没有作者书面许可不得转载。
推荐链接
|
| 返回顶楼 | |
