R: Dynamic Time Warping Barycenter Averaging

dba {IncDTW}

R Documentation

Dynamic Time Warping Barycenter Averaging

Description

Average multiple time series that are non-linearly aligned by Dynamic Time Warping. Find the centroid of a list of time series.

Usage

dba(lot, m0 = NULL, iterMax = 10, eps = NULL, 
                dist_method = c("norm1", "norm2", "norm2_square"),
                step_pattern = c("symmetric2", "symmetric1"),
                ws = NULL,
                iter_dist_method = c("dtw_norm1", "dtw_norm2",
                                     "norm1","norm2", "max", "min"), 
                plotit = FALSE)
                
# deprecated
DBA(lot, m0 = NULL, iterMax = 10, eps = NULL, 
                dist_method = c("norm1", "norm2", "norm2_square"),
                step_pattern = c("symmetric2", "symmetric1"),
                ws = NULL,
                iter_dist_method = c("dtw_norm1", "dtw_norm2",
                                     "norm1","norm2", "max", "min"), 
                plotit = FALSE)
                
centroid(lot, dist_method = c("norm1", "norm2", "norm2_square"),
              step_pattern = c("symmetric2", "symmetric1"),
              normalize = TRUE, ws = NULL, ncores = NULL,
              useRcppParallel = TRUE)


              
## S3 method for class 'dba'
print(x, digits = getOption("digits"),  ...)

## S3 method for class 'dba'
summary(object, ...)              

is.dba(x)

Arguments

`lot`	List of time series. Each entry of the list is a time series as described in `dtw2vec`.
`m0`	time series as vector or matrix. If `m0` is `NULL`, the initial time series `m0` is determined by `centroid` as the centroid of `lot`, which is the one time series of `lot` with the minimum average DTW distance to all other time series of `lot`.
`iterMax`	integer, number of maximum iterations
`eps`	numeric, threshold parameter that causes the algorithm to break if the distance of two consecutive barycenters are closer than eps
`dist_method`	character, describes the method of distance measure. See also `dtw`.
`step_pattern`	character, describes the step pattern. See also `dtw`.
`ws`	integer, describes the window size for the sakoe chiba window. If NULL, then no window is applied. (default = NULL)
`iter_dist_method`	character, that describes how the distance between two consecutive barycenter iterations are defined (default = "dtw")
`plotit`	logical, if the iterations should be plotted or not (only possible for univariate time series)
`normalize`	logical, default is TRUE, passed to `dtw_dismat`
`ncores`	integer, default = NULL, passed to `dtw_dismat`
`useRcppParallel`	logical, default is TRUE, passed to `dtw_dismat`
`x`	output from `dba`
`object`	any R object
`digits`	passed to `round` and `print`
`...`	additional arguments, e.g. passed to `print` or `summary`

Details

The parameter iter_dist_method describes the method to measure the progress between two iterations. For two consecutive centroid candidates m1 and m2 the following methods are implemented:

'dtw_norm1': dtw2vec(m1, m2, dist_method = "norm1", step_pattern = "symmetric2")$normalized_distance

'idm_dtw2': dtw2vec(m1, m2, dist_method = "norm2", step_pattern = "symmetric2")$normalized_distance

'idm_norm1': sum(abs(m1-m2))/(ncol(m1) * 2 * nrow(m1))

'idm_norm2': sqrt(sum((m1-m2)^2))/(ncol(m1) * 2 * nrow(m1))

'idm_max': max(abs(m1-m2))

'idm_min': min(abs(m1-m2))

Value

`call`	function call
`m1`	new centroid/ bary center of the list of time series
`iterations`	list of time series that are the best centroid of the respective iteration
`iterDist_m2lot`	list of distances of the iterations to lot
`iterDist_m2lot_norm`	list of normalized distances of the iterations to lot
`iterDist_m2m`	vector of distances of the iterations to their ancestors
`centroid_index`	integer giving the index of the centroid time series of `lot`
`dismat_result`	list of results of `dtw_dismat` called by `centroid`

References

Leodolter, M.; Pland, C.; Brändle, N; IncDTW: An R Package for Incremental Calculation of Dynamic Time Warping. Journal of Statistical Software, 99(9), 1-23. doi: 10.18637/jss.v099.i09
Sakoe, H.; Chiba, S., Dynamic programming algorithm optimization for spoken word recognition, Acoustics, Speech, and Signal Processing [see also IEEE Transactions on Signal Processing], IEEE Transactions on , vol.26, no.1, pp. 43-49, Feb 1978. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1163055
Petitjean, F; Ketterlin, A; Gancarski, P, A global averaging method for dynamic time warping, with applications to clustering, Pattern Recognition, Volume 44, Issue 3, 2011, Pages 678-693, ISSN 0031-3203

Examples

## Not run: 
data("drink_glass")
# initialize with any time series 
m1 <- dba(lot = drink_glass[1:10], m0 = drink_glass[[1]], 
          dist_method = "norm2", iterMax = 20)

# initialize with the centroid

tmp <- centroid(drink_glass)
cent <- drink_glass[[tmp$centroid_index]]
m1 <- dba(lot = drink_glass[1:10], m0 = cent, 
          dist_method = "norm2", iterMax = 20)

# plot all dimensions of the barycenters m_n per iteration:
plot(m1)

# plot the distances of the barycenter of one iteration m_n
# to the barycenter of the previous iteration m_n-1:
plot(m1, type = "m2m")

# plot the average distances of the barycenter m_n 
# to the list of time series:
plot(m1, type = "m2lot")


## End(Not run)

[Package IncDTW version 1.1.4.4 Index]