Discussion questions#
kNN makes predictions based on similarity between observations rather than estimating explicit parameters like regression models do. When might this way of thinking about prediction feel useful, and when might a regression approach feel more appropriate?
kNN relies heavily on distance in feature space to determine similarity of observsations. In psychological or neuroscience data, what does “similarity” actually mean? Who decides what counts as similar? When might distance be a meaningful representation of reality, and when might it be misleading?