Everything you know about Dynamic Time Warping is Wrong

412 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Semi-supervised time series classification (2006, methodology): “All the data records have been normalized and rescaled to have length 85 (recent results suggest that we lose nothing by rescaling [35]).”
Learning-Based Methods for Comparing Sequences, with Applications to Audio-to-MIDI Alignment and Matching (2016, background): “In data mining applications (Ratanamahatana and Keogh, 2004) it is argued that the band radius path constraint both reduces computational complexity and results in more reliable alignments by avoiding paths with many non-diagonal moves.”
Gesture recognition for human-robot collaboration: A review (2017, unknown): “er, 1982) Time and space complexity (Ratanamahatana and Keogh, 2004)”
Highly Comparative Feature-Based Time-Series Classification (2014, unknown): “2.3.”
Using dynamic time warping distances as features for improved time series classification (2016, background): “This window constraint not only speeds up computation of DTW, but somewhat non-intuitively, also improves the accuracy of time series classification (Ratanamahatana and Keogh 2004a).”
Generalizing DTW to the multi-dimensional case requires an adaptive approach (2016, unknown): “For small datasets, DTW and ED often make different decisions about which item is the nearest neighbor, but as the datasets get larger they tend to agree more and more often, eventually converging to the same error-rate (Shieh and Keogh 2009; Ratanamahatana and Keogh 2004).”