How can I get un-descend index of latent in pca function

Hi, I'm trying to practice Neuron Spike Sorting using pca and kmeans clustering.

When I plot latent from pca function, I only get latent values which is desecended.

I want to use only 3 of them that has most powerful influence..

However, I can't find the dimension index of the big three.

My final result of spikes sorting is queit not good. I put k=2 in kmenas, but when I plot those two group, it looks same.

I guess I have a mistake finding big three powerful dimension in raw data.

Please,

spikes has 80000x48 double.

Here is my code :

%Spike Sorting
load('spikes.mat'); %detected spikes
[coeff,score,latent,tsquared,explained,mu] = pca(spikes); %PCA analysis

nscore1 = normalize(score(:,1));
nscore2 = normalize(score(:,2));  %score data normalize

nscore(:,1) = nscore1;
nscore(:,2) = nscore2;

[idx,C] = kmeans(nscore,2); %idx is clustering index

Sidx = [spikes, idx];

for k=1:size(idx)
if(Sidx(k,2)==1)
 figure(1)
 title('cluster1 spikes')
 plot(spikes(k,:))
 hold on
end
end

for k=1:size(idx)
if(Sidx(k,2)==2)
 figure(2)
 title('cluster2 spikes')
 plot(spikes(k,:))
 hold on
end
end

load('spikes.mat'); %detected spikes
[coeff,score,latent,tsquared,explained,mu] = pca(spikes); %PCA analysis

nscore1 = normalize(score(:,1));
nscore2 = normalize(score(:,2));
nscore3 = normalize(score(:,47));    


figure(1)
scatter(nscore1, nscore2);
title('Normalize score1 vs Normalize score2')
xlabel('Normalize score1')
ylabel('Normalize score2')

figure(2)
scatter(nscore1, nscore3);
title('Normalize score1 vs Normalize score3')
xlabel('Normalize score1')
ylabel('Normalize score3')

figure(3)
scatter(nscore2, nscore3)
title('Normalize score2 vs Normalize score3')
xlabel('Normalize score2')
ylabel('Normalize score3')

figure(4)
scatter3(nscore1, nscore2, nscore3)
title('Normalize score1,2,3')
xlabel('Normalize score1')
ylabel('Normalize score2')
zlabel('Normalize score3')

figure(5)
plot(latent, 'o')
title('principal component') %eigen plotting

Answer 1

[the cyclist]

My answer to this question gives a detailed explanation of how to use MATLAB's pca function, including how to reduce the number of dimensions. Please read the question, my answer, the other users' comments, and my responses to those. I think it will help you.

[WonJun Choi]

wow, I think I understand what I have to do. But before I go, can I double check whether I really understand?

So when i use pca and get 'score, latent', these two datas are already descended according to variance, the first dimension of data contains the most powerful tendency in whole data dimension.

Do I understand correct?

[the cyclist]

Correct.

Just to be clear, you want all rows of the first 3 columns of score (so that you are using all observations of the PC features that contribute the most variation).

The sum of the first 3 values of explained will tell you what fraction of the original variation you have retained.

[WonJun Choi]

Thank you for your adivce! Hope you have great day.