choice.keepX from tune function: minimize 'cor' measure?

Analysis
1

Hello everyone,

I have a little question concerning the ‘tune’ function.
Indeed, in sPLS Liver Toxicity Case Study | mixOmics, it is written that for PLS2, we should use ‘cor’ instead of ‘MAE’ measure. By doing this and plotting the result, I get this picture.

tune_corMeasure_PLS2
tune_corMeasure_PLS2875×540 10.1 KB

However, by asking tune.spls$choice.keepX, I get
tune_corMeasure_PLS2_2

Hence, it seems choice.keepX minimizes the measure. Shouldn’t we maximize it instead, as we want the highest correlation between predicted and actual components?

Best regards,
Emile M.

2

hi @emile.mardoc

I agree with you that you should maximise the correlation and as shown in the link you mentioned you should see some dot plots instead of the line plots. I wonder if you called the tune function incorrectly?

If you can send me the code (e.g on the liver.toxicity data, pretending this is the same structure of the data) I can have a quick look.

Kim-Anh