Truncated PLS

With the simple assumption of noise being normally distributed, we make an extra step in the NIPALS algorithm for PLS that truncates or down-weights values in loadings (or scores) that seem to originate from normal or t-distributions. For spectrometric data or other data with zeros baselines this can simplify interpretations and hopefully help in the "search for the needle in the heystack" in megavariate data analysis. Some theory, implementations and practical examples will be discussed.