Temporal Tensor Decomposition, a Dimensionality Reduction Tool for Longitudinal Multivariate Data
Calculate the de-noised temporal tensor
Aggregate features using feature loadings
Caculate the Bernoulli kernel
Estimate subject loading of testing data
Format data table into the input of tempted
Plot nonparametric smoothed mean and error bands of features versus ti...
Plot nonparametric smoothed mesan and error bands of meta features ver...
Plot the temporal loading functions
Take log ratio of the abundance of top features over bottom features
Reconstruct tensor from low dimensional components
Remove the mean structure of the temporal tensor
Decomposition of temporal tensor
Run all major functions of tempted
TEMPoral TEnsor Decomposition (TEMPTED), is a dimension reduction method for multivariate longitudinal data with varying temporal sampling. It formats the data into a temporal tensor and decomposes it into a summation of low-dimensional components, each consisting of a subject loading vector, a feature loading vector, and a continuous temporal loading function. These loadings provide a low-dimensional representation of subjects or samples and can be used to identify features associated with clusters of subjects or samples. TEMPTED provides the flexibility of allowing subjects to have different temporal sampling, so time points do not need to be binned, and missing time points do not need to be imputed.