plot PCA projection: How interpret



I want to perform PCA on my data_array object.
How do I get a nice plot of the data projected onto the first two principal components?

I tried

using ManifoldLearning
using Plots

M1 = fit(PCA, data_array; maxoutdim=2)
# transforms observations data_array into PCs
Y1 = transform(M1, data_array) # 2x547
X_PCA = reconstruct(M1, Y1) # 96x547
M1_proj = projection(M1) # 96x2

Plots.scatter(M1_proj, title="PCA projection", legend=false)

but I cannot quite interpret the plot… It does not look any like standard PCA projection plots.

Any ideas?


you are plotting the rotation matrix, which is usually plotted as arrows in a biplot (see here for an example), you have to plot Y1 for the actual points.


Thanks! I just realised it is Y1. However, how can I plot the rotation matrix, or rather the new directions in the old coordinate system ?


You use the rotation matrix as coordinates for the arrow tips, all arrows start from the origin. You may have to scale them, so they fit nicely into your data.


Thanks, yes, I know. That is my question. I can’t find the right package/function.


For serious plotting I still use RCall, depending on where you come from you might want to consider PyPlot too.