数学建模竞赛蠓虫分类.docx
《数学建模竞赛蠓虫分类.docx》由会员分享,可在线阅读,更多相关《数学建模竞赛蠓虫分类.docx(8页珍藏版)》请在冰豆网上搜索。
![数学建模竞赛蠓虫分类.docx](https://file1.bdocx.com/fileroot1/2023-7/8/667d8813-0de3-4af8-8f46-30f9ed8b5d65/667d8813-0de3-4af8-8f46-30f9ed8b5d651.gif)
数学建模竞赛蠓虫分类
解:
(1)BayeS判别:
求条件概率
假设假设两类数据均服从二维正态分布,冃1〜NLD=〜N「"2)
P(I)=Pe2p0.5
则条件概率为:
A1
4j=Nzxk
NiX^^i
A1AA
Li=Mς(Xk一μi)(x^μi)
NiXk知
PeilX)=2P(XZP(J)
ςP(XPj)PCj)
円
决策规则为;
如果Pwi|x)=maxp@j∣x),则XE皎i
ij=1,2ji
或等价地:
P(Xwi)Pwi)=maxP(Xj)P3j),则X
j=1,2JJ
I(X)=
P(XlI)P
(2)p(χ∣2)P(I)
判决函数:
1d1
gi(x)=「2&一Ii)^i^1(^lib-∣n^--In^iPInP(
决策面方程为gi(χ)=Igj(X)即
冷[(x-S)=(x」i)-(x-Jj)^rI(^lj)ΓInp^)
类似地,BayeS最小风险判别可通过给出风险后得到。
x=[1.201.301.181.141.261.281.361.48
1.40
1.381.241.38
1.54
1.38
1.56
1.24
1.28
1.401.221.36;
1.861.961.781.78
2.00
2.00
1.74
1.82
1.70
1.901.721.64
1.82
1.82
2.08
1.80
1.782.041.881.78];
n1=6;n2=9;n3=5;
plot(x(1,1:
n1),x(2,1:
n1),'o',x(1,n1+1:
n1+n2),x(2,n1+1:
n1+n2),'*',x(1,n1+n2+1:
end),x(2,n1+n2+1:
end),'r+');
mm1=sum(y(1:
n1))/n1;
mm2=sum(y(n1+1:
n1+n2))/n2;
sgm1=cov(x(:
1:
n1)');%=s1/(n1-1);
sgm2=cov(x(:
n1+1:
n1+n2)');
X=x(:
n1+n2+1:
end);
X=X';
pxw1=mvnpdf(X,m1',sgm1);
pxw2=mvnpdf(X,m2',sgm2);
pwx1=pxw1./(pxw1+pxw2);
pwx2=pxw2./(pxw1+pxw2);
display('UsingBayesprincipalis:
')
Apf=find(pwx1>pwx2)+n1+n2,
(2)FiSher判别:
求投影方向W
圉4.3FiSher线性判别的基本原理
准则函数:
其中
最优解:
JF(W)=
(ff∣1-r~2)2
S1S2
mi
1
Ni
WTSbW
WTsW
mi
NiyYi
y,
(x-mj(x-mj)τ,i=1,2
S八
XWyi
S=SIs2
Si
S.
Yn
if
X(y-mi)2,
Y-Yi
SiS2
S1(m1m2)
(1)
Yo
m1m2
2
NlmIN2ιτi2
N1N2
(3)
Yo
_m1m2In(P
2N
m1=mean(x(:
1:
n1),2);m2=mean(x(:
n1+1:
n1+n2),2);s1=(x(:
1:
n1)-repmat(m1,1,n1))*(x(:
1:
n1)-repmat(m1,1,n1))';
s2=(x(:
n1+1:
n1+n2)-repmat(m2,1,n2))*(x(:
n1+1:
n1+n2)-repmat(m2,1,n2))';
S=s1+s2;w=inv(S)*(m1-m2);y=w'*x;
mm1=sum(y(1:
n1))/n1;mm2=sum(y(n1+1:
n1+n2))/n2;y0=(mm1+mm2)/2;
%y0=(mm1*n1+mm2*n2)/(n1+n2);dpyb=y(n1+n2+1:
end);
display('Usingfisherprincipalis:
')Apf=find(dpyb>y0)+n1+n2,figure
(2);
t=1.1:
0.01:
1.6;
kkk=-w
(1)∕w
(2);
ft=kkk*t+yθ∕w
(2);
PIot(X(1,1:
n1),x(2,1:
n1),'o',x(1,n1+1:
n1+n2),x(2,n1+1:
n1+n2),'*',x(1,n1+n2+1:
end),x(2,n1+n2+1:
end),'r+',t,ft);
axis([1.1,1.6,1.621]);
感知器准则及梯度下降算法:
TynO,nT,2,,N
(b)规范化
(a)耒规范化
图4.5
解区和解向堆示意图
梯度下降法:
批处理感知器算法
bigin
retUrn
end
initioaliz
do
Unt
il
固定增量单样本感知器
1.
bigin
initioaliz
:
ea,
2
do
k∙k
3
k
ify
被a错
4
—Until
所有模
5
return
a
6
end
%peceptron
x1=[ones(1,length(x));x];
x1(:
n1+1:
n1+n2)=-x1(:
n1+1:
n1+n2);
epsl=0.1;
a=1-2*rand(3,1);
k=0;
whilek<100000
k=k+1;
y=x1(:
rem(k,n1+n2)+1);
ifa'*ya=a+y;
end
end
y1=a'*x1(:
1:
n1+n2);
ind=find(y1<=0);
display('theSamPIeSforfirstclassUSing
PeCePtrOnPrinCiPaIis:
')
Apf=find(a'*x1(:
n1+n2+1:
end)>0)
Pt=-a
(2)*"a(3)-a
(1)∕a(3);
figure(3),
PlOt(X(1,1:
n1),x(2,1:
n1),'o',x(1,n1+1:
n1+n2),x(2,n
1+1:
n1+n2),'*',x(1,n1+n2+1:
end),x(2,n1+n2+1:
end),'r+',t,[pt,ft]);
最小平方误差准则:
设
Ya=
b
y1T
y11
II
y12
y1
T
Y=∣y2
I:
I
11
IIy21
I=I...
y22
%
…I
-yN
11
ILyNI
yN2
IyN(?
bNC
N
目标:
minJs(a)=IleP=|Ya-b∣∣2=瓦(aτyn-bn)2(463)
N
7s(a)八2(aTy^bn)y^2Yt(Ya-b)(A-64)
n=1
n=1
目标函数的梯度:
VJs(aPO^=>YTYa=YTb(4-65)
Aa=(YTY)1YTb=Yb(4
%MSE
b=ones(n1+n2,1);
Y=x1(:
1:
n1+n2);
Y=Y';
Yplus=(Y'*Y)^(-1)*Y';ahat=Yplus*b;mt=-ahat
(2)*t/ahat(3)-ahat
(1)/ahat(3);
figure(4)
plot(t,[pt;ft;mt],x(1,1:
n1),x(2,1:
n1),'o',x(1,n1+1:
n1+n2),x(2,n1+1:
n1+n2),'*',x(1,n1+n2+1:
end),x(2,n1+n2+1:
end),'r+');
axis([1.1,1.6,1.6,2.1]);display('UsingMSEprincipalis:
')Apf=find(ahat'*x1(:
n1+n2+1:
end)>0)+n1+n2,