Hi all, I would like to know how to calculate PRESS and TSS for my PLSDA. The main reason is I would like to follow the methodology in this paper (PDF) Discriminant Q 2 (DQ 2 ) for improved discrimination in PLSDA models
My first issue is i don’t know the meaning of the predicted values (from predict function) - I have tried multiplying the reg.coefficients (from predict(plsdamodel, X)$B.hat) by the centred and scaled X matrix (plsdamodel$X), but my predicted values don’t match the predicted values I obtain from predict(plsdamodel, X), and I think I remember reading somewhere that the predict(plsdamodel,X)$predict would give me the likelihood of belonging to some class - so if this is true, it’s expectable that my values don’t match, but also I can’t really use likelihood to calculate the TSS and PRESS. I need some sort of predicted value that leads to a predicted class, depending on some threshold.
Note: I know PRESS is calculated based on cross-validation. I am just starting by the simplest example to understand where to get the data before going into the perf with CV loo, but the ultimate goal is to 1- get predicted values to be able to 2- calculate PRESS 3- maybe even calculate PRESSD if I manage to code it in R and then 4-get the DQ2 from there, otherwise 2- calculate PRESS and 3- get Q2.
Let me know if I need to clarify this question a bit, I can also create a set of fake data and do the codes if it’s easier.