fix VL bug, and use lag to shift local variables so for every trial we only consider previous local params.

remove current trial from local stat calculations.

init a new dummy var analysis for ccn2019

pls and RMSE for accuracy

refactor modeling "correct" to another script

a new chunk to mark and remove highly correlated predictors for glm, etc

rewrite splitting into train/test so both will be a representative samples of the original "correct" ratio

comments and PLS VarImp diagrams

add variable importance for RT and accuracy

add PLS model for RT

plot PLS model output

two models, one for commonly used model of the n-back, and a new model to contain all local and global vars.

caret, AUC, etc for CCN2019