.exe file steps and settings for CCA to use running in R

[aelliott08]

Hi,

Having some trouble using the .exe file but I found a CCA package for R that looks like its straight from the matlab toolbox -- “ProjectionBasedClustering” package, CCA function. (Would just use the .exe but I cant seem to get the correct matlab + compiler). Should be directly transferable but I cant seem to find the steps taken in the .exe file or the settings used for CCA. Paper just refers to the .exe program - which I cant currently access.

Any chance I could get info on the steps taken in the .exe program or the settings used for CCA? With that I could run this all in R and on mac. Thanks!

[daniellyz]

Dear alliott08,

Thanks a lot for testing MetICA.

Before using CCA for cluster number determination, you should install it properly by downloading the exe file inside: https://github.com/daniellyz/MetICA/tree/master/MetICA_CCA/for_redistribution

Through installation, you will be directed to install Matlab Runtime Compiler which is mendatory to run MetICA_CCA.exe. The installation will take a few minutes. Then you could download the entire folder and run MetICA_CCA.exe a graphical interface for cluster visualization. The program now only works on Windows.

I admit that it is not straightforward to jump from R to this graphical interface and back to R... Now we have developped a Beta-version R package which can be found at:
https://github.com/daniellyz/MetICA2
You can find inside the tutorial of the package. You will see that all steps are performed in R (2-3 command lines). We suggest 6 plots to determine the number of clusters. It will be a great pleasure for us if you could test the 2nd version of MetICA!

Thanks a again and have a nice weekend!

[aelliott08]

Is there a record anywhere of the exact preprocessing steps included in the .exe file?

[aelliott08]

Actually I found your pipeline very easy to use!

Would you be willing to release the matlab code? (vs the exe file)

I cannot access matlab/compiler because my institution will not allow it at present. I’m really interested in creating the CCA plot mentioned in the tutorial (and used in the paper) . I would like to replicate your CCA plot per tutorial. When recoding the CCA from Matlab toolbox in R, I cannot discover from existing documentation:
- What are you using for Epochs
- Why are you reading in the source matrix?
- Your tuning parameters.

Not sure if thats documented anywhere? Or if you would share the matlab code?

[daniellyz]

Dear alliott08,

Thanks again for your interest in MetICA. I'm happy that you think the utilization is straight forward!!
The core function for CCA analysis in the GUI can be found at:
http://www.cis.hut.fi/Opinnot/T-61.3030/harjtyo/mycca.m
The way I used this function is:
P=cca(S',2,200,D);
where S is the source matrix (source_list.txt), 2 means we project the estimated components on a two-dimension space and 200 is the training length Epoch, D is the distance matrix (distance.txt).
The P will be the projected scores of these estimates on the CCA space (similar to PCA scores).

CCA is now also available in R as you mentioned in the ProjectionBasedClustering package.
https://rdrr.io/cran/ProjectionBasedClustering/man/CCA.html
It is directly translated from the Matlab package. So I guess it will work the similar way as in Matlab. Only in this case the source matrix S is not necessary! You can put directly the distance matrix:
P=CCA(D,200,OutputDimension=2)

Then for the plot, if you want to check the situation of 9 clusters, you need to first import "cluster_label.txt" as matrix (This is matlab code but you can do it in R of course)
nbc = 9;
cluster_labels = C(:, nbc -1); % The cluster label of each estimate.
hold on
for c= 1:nbc,
f=find(cluster_labels==c)
plot(P(f,1),P(f,2),'color',rand(1,3)) % Plot each cluster with a random color
end
hold off

I will compare the output of R package with Matlab on my dataset as well! In the end the idea is to incoporate this visualization in the 2nd version of MetICA

Thank you again. Your experiences has helped us improve this package!
Best regards
Youzhong