Matlab中PCA实现3D转2D：princomp返回3D矩阵的问题咨询

Hey there! Let's sort out why you're getting a 3D matrix instead of the 2D output you need for plotting. Here's what's going on and how to fix it:

The princomp function (a legacy MATLAB tool now) returns scores for all principal components by default. Your input data' is a 3x3 matrix (3 samples, each with 3 features), so the output x ends up being a 3x3 matrix—each column represents a principal component, and each row maps to one of your samples. That's why you're seeing a 3-dimensional structure instead of the 2D you want.

You just need to extract the first two columns of the scores matrix—these are the top two principal components that capture the most variance in your data. Adjust your code like this:

data = [5 4 5; 5 3 0; 1 2 2];
% Transpose because princomp expects each column to be a sample
[coeff, score, latent] = princomp(data');
% Grab the first 2 columns to get your 2D representation
x_2d = score(:, 1:2);

Now x_2d is a 3x2 matrix, exactly what you need for plotting your 3 samples in 2D space.

Since princomp is deprecated, MATLAB recommends using the pca function instead. It's more straightforward, and you can directly specify how many components you want:

data = [5 4 5; 5 3 0; 1 2 2];
% pca expects each row to be a sample—no transpose needed!
% Ask for exactly 2 principal components
[x_2d, coeff, latent] = pca(data, 'NumComponents', 2);

This code will spit out a 3x2 matrix x_2d right away, ready for your plots. A quick note: pca uses rows as samples by default, which is the opposite of princomp—that's a common pitfall, so this saves you from having to remember to transpose!

If you want to confirm how much of your data's variance is retained in the 2D space, take a look at the latent vector. The values represent the variance explained by each principal component—sum the first two and divide by the total sum to get the percentage of variance you're keeping.

内容的提问来源于stack exchange，提问作者tamaramaria