paranoia / PiecewiseLinear / 0.1.2

Split a multidimensional timeseries such that a piece linear reconstruction has minimal error.

The input is first normalized such that each timeseries has a mean of 0 and a standard deviation of 0. This is to avoid overweighting one of the dimensions vs the others simply because of scaling differences.

The resulting timeseries is then run through a dynamic programming algorithm that produces the split with minimal SSE, running in time proportional to the size of the timeseries input and the number of segments.