Timeseries Model¶

The Time Series model was originally developed to evaluate similarities between waveforms generated by electrical circuit simulations. However, this model can be used more generally to compare any set of time series data, so long as the starting and ending times for each sequence of values are the same. Although the time series inputs do not need to be sampled identically, our initial step is to resample since the analysis ultimately requires corresponding samples for comparison. Points in each sequence are binned, with the bin size calculated by dividing the time range into equal intervals. The values of the points within each bin are then averaged. The underlying assumption is that there are enough samples in the sequence to have at least one sample per bin. The number of bins is a user-supplied parameter to the analysis.

Slycatâ„¢ calculates a table of the distances between each pair of resampled time series by summing the differences between corresponding points. The distance table is used to create an agglomerative, hierarchical model of similarities between the sequences. The resulting model is a tree, in which the set of time series contained within each subtree are more and more similar as successive subtrees get closer to the leaves. Because our distance metric was developed for electrical simulation data, both amplitude (y value) and timing (x value) must match for a pair of sequences to be regarded as similar (i.e. identical shapes that are shifted in time are not seen as similar).