The size of my PCA isn't correct

5 次查看（过去 30 天）
Héctor
Héctor 2016-1-18
评论： Frank Nylander 2024-2-4
I am trying to perform principal component analysis using pca and not princomp. My dataset consists of 303 samples each containing 3904 dimensions, which explains why I want to perform PCA. My data is in a matrix called Data which has a size of 303 (rows) by 3904 (columns). When I execute the PCA command with
[coeff,score] = pca(Data);
I am returned a matrix coeff of size 3904*302 and the matrix score is of size 303*302 which doesn't seem to be correct right? Online documentation suggests that coeff should be 3904*3904 (i.e. square with the dimension size of the variables, here the columns). What is going wrong here?
ps: I also tried transposing the data and this results in a coeff matrix of 303*303

请先登录，再回答此问题。

回答（1 个）

Roger Wohlwend
Roger Wohlwend 2016-1-19
It seems that Matlab expects a Matrix with more observations than variables (i.e. more rows than columns). As you violate that rule (only 303 observations for 3904 variables), Matlab transposes the Matrix in order to obtain a Matrix with more observations than variables. If you transpose the Matrix, Matlab interprets your variables as observations and the other way around. So the result is not valid. You cannot solve this Problem. If you don't have more than 303 observations, your Matrix must not have mor than 303 columns.

类别

在 Help Center 和 File Exchange 中查找有关 Dimensionality Reduction and Feature Extraction 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by