Meaning of 'Data Adaptive' and 'Non Data Adaptive' time series representations I was reading a paper and came across a tree-figure which split time series representations into "Data Adaptive" and "Non Data Adaptive" representations. In the Data Adaptive branch were representations like SVD, Sorted coefficients, piecewise polynomial etc. While in the "Non Data Adaptive" branch were representations wuch as wavelets, fourier/cosine transforms, random mappings etc.
So I was wondering, what is meant by 'Data Adaptive' and 'Non Data Adaptive'?
 A: The paper you most likely read came from the work of Keogh and Lin et al and possibly talked about Symbolic Aggregate Approximation SAX
The best answer i can give is to cite a paper below:
"Data Adaptive representations: in this category, a common representation will be 
chosen for all items in the database that minimizes the global reconstruction error".
–
"Non-Data Adaptive
representations: in contrast, these methods consider local properties of the data, and construct an approximate representation accordingly"
Experimental Comparison of Representation Methods and
Distance Measures for Time Series Data
XiaoyueWang · Hui Ding · Goce Trajcevski ·
Peter Scheuermann · Eamonn Keogh
In addition in the Master-Thesis von Irina Alles - September 2013
https://team.inria.fr/zenith/files/2013/11/ia_ma_thesis_final.pdf
Complete descriptions with examples are given :-)
Non data-adaptive representation - Page 14
Non data-adaptive techniques use the same set of parameters for dimensionality reduction regardless of the underlying data. One of the early works on this topic was achieved by Agrawal et al. [1993], who used the Discrete Fourier Transform (DFT) ..."
"Data-adaptive representation - Page 15
This category of time series representations assembles techniques which take into account the underlying data and adjust their parameters accordingly. Almost any non-data adaptive approach can become data adaptive by adding a parameter selection step [Esling and Agon, 2012]. Vlachos et al. [2004],Struzik and Siebes [1999] realised this idea for DFT and DWT...."
