How can I use fitcdiscr using my own covariance matrix?

Question

Johannes 2018-1-25

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/378923-how-can-i-use-fitcdiscr-using-my-own-covariance-matrix

评论： Johannes 2018-2-1

Hi, I want to train a classifier using my own covariance matrix and/or use data that are already whitened (using a covariance matrix that was estimated from more data than the data that are fed into the classifier). However, I do not see any option to supply a covariance matrix to fitcsdiscr. Is there a way to do this? There is a function called makecdiscr that allows one to supply a covariance matrix, but this seems to create a classifier without fitting/training? Or is the outcome of makecdiscr the same as that of fitcsdiscr, just based on different inputs?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Bernhard Suhm 2018-2-1

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/378923-how-can-i-use-fitcdiscr-using-my-own-covariance-matrix#answer_303056

The training of fitcdiscr is to estimate the covariance matrix. If you have the covariance matrix, makecdiscr can be used to generate a discriminant analysis classifier. The output is somewhat different (only what we call a "compact" model from makecdiscr), because the original data is not passed to makediscr. The output of the fit functions includes a reference to the training data.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Johannes 2018-2-1

Ah OK, silly me! So in short, makecdiscr can produce a very similar classifier as fitcdiscr provided that the supplied covariance matrix is similar to the one that would otherwise be estimated by fitcsdiscr. Clear answer, thanks.

请先登录，再进行评论。