Transformer (Pairwise Panel)

DtwDtaidistMultiv

Multivariate Unequal length Missing values

Multivariate dynamic time warping distance, from dtaidistance.

Direct interface to dtaidistance.dtw_ndim.distance_matrix and dtaidistance.dtw_ndim.distance_matrix_fast.

This distance is covers multivariate data. For univariate data and the default euclidean distance, DtwDtaidistUniv is mathematically equivalent but may be more efficient.

To specify the internal distance to be separate from squared euclidean, use AlignerDtwDtai inside a DistFromAligner.

Schnellstart

python

from sktime.dists_kernels.dtw import DtwDtaidistMultiv

estimator = DtwDtaidistMultiv(use_c=False, window=None, max_dist=None, max_step=None, max_length_diff=None, penalty=None, psi=None, use_pruning=False)

Parameter(8)

use_c: bool, optional, default=False: Whether to use the faster C variant: True for C, False for Python. True requires a C compiled installation of dtaidistance.
If False, uses dtaidistance.dtw_ndim.distance_matrix.
If True, uses dtaidistance.dtw_ndim.distance_matrix_fast.
windowinteger, optional, default=infinite: Sakoe Chiba window width, from diagonal to boundary. Only allow for maximal shifts from the two diagonals smaller than this number. The maximally allowed warping, thus difference between indices i in series 1 and j in series 2, is thus |i-j| < 2*window + |len(s1) - len(s2)|. It includes the diagonal, meaning that Euclidean distance is obtained by setting window=1. If the two series are of equal length, this means that the band appearing on the cumulative cost matrix is of width 2*window-1. In other definitions of DTW this number may be referred to as the window instead.
max_dist: float, optional, default=infinite: Stop if the returned values will be larger than this value.
max_step: float, optional, default=infinite: Do not allow steps larger than this value. If the difference between two values in the two series is larger than this, thus if |s1[i]-s2[j]| > max_step, replace that value with infinity.
max_length_diff: int, optional, default=infinite: Return infinity if difference of length of two series is larger than this value.
penalty: float, optional, default=0: Penalty to add if compression or expansion is applied
psi: integer or 4-tuple of integers or none, optional, default=none: Psi relaxation parameter (ignore start and end of matching). If psi is a single integer, it is used for both start and end relaxations for both series in a pair of series. If psi is a 4-tuple, it is used as the psi-relaxation for (begin series1, end series1, begin series2, end series2). Useful for cyclical series.
use_pruning: bool, optional, default=False: Prune values based on Euclidean distance.

Referenzen

[1]

H. Sakoe, S. Chiba, “Dynamic programming algorithm optimization for spoken word recognition,” IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 26(1), pp. 43–49, 1978.