PTA3 on fMRI with time penalisations

The same analysis using a identity metric but penalisation for time -the time components (all) are asked to be smooth : the smoother is a non-linear kernel smoother i.e. the weightings depends on the distance of points (traditional bandwidth) and distance of values (preserving some high peaks)- gives for the first Principal tensor:

summary of the decomposition

The first Principal Tensor is very similar to the "canonical" analysis, with a slightly lower singular value as one constraints the time component to lie in a subspace of the L_2. Notheless note for this component that one F has been classed as outlier from the boxplot procedure ten expressing a more homogeneous female group (apart this outlier).

some differences can also be seen for example for te first PT associated to te subject component