作者: chungyuandye (養花種魚數月亮賞星星)
標題: Re: [問題]Covariance不等,做discriminant analysis
時間: Tue Dec 4 06:51:15 2012
※ 引述《agnes12 (agnes)》之銘言:
: Hi~
: 謝謝您的回應:)
: 我發現我的課本用的名詞有點confusing,
: 我用的課本的discriminant analysis是指canonical discriminant analysis
: 而一般(我先前google時發現)discriminant analysis
: 在我的課本用的名詞是classification analysis.
: 我現在還是疑惑我能不能做canonical discriminant analysis
: 因為data is of unequal covariance matrice ><
: ※ 引述《chungyuandye (養花種魚數月亮賞星星)》之銘言:
: : http://tinyurl.com/c5su2np
: : Quadratic Discriminant Analysis for SPSS.
I think you can analyze it with Fisher's canonical discriminant analysis/
If you try Fisher's original iris dataset, you can find that
the Covariance Matrices are Unequal.
http://tinyurl.com/ckflxg9
Discrimination methods are very much affected by the nature of
the within-group covariance matrices. The covariance matrices
are assumed equal for allgroups in the Fisher, or linear discrimination
methodology. When the assumption of homogeneity of within-groups
covariance is grossly violated, quadratic discriminant analysis is
recommended.
http://www.statistical-solutions-software.com/BMDP-documents/BMDP-5M.pdf
Linear discriminant analysis (LDA) is one approach to dimensionality reduction
that makes use of a linear transformation matrix. The widely used Fisher's
LDA is "sub-optimal" when the sample class covariance matrices are unequal,
meaning that another linear transformation exists that produces lower loss in
discrimination power.
http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=940332
--
養花種魚數月亮賞星星
http://chungyuandye.twbbs.org
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 106.1.133.106
沒有留言:
張貼留言