IncDTW (version 1.1.4.5)

dtw_dismat: DTW Distance Matrix/ Distance Vector

Description

Calculate a matrix of pairwise DTW distances for a set of univariate or multivariate time series. The output matrix (or dist object) of DTW distances can easily be applied for clustering the set of time series. Or calculate a vector of DTW distances of a set of time series all relative to one query time series. Parallel computations are possible.

Usage

dtw_dismat(lot, dist_method = c("norm1", "norm2", "norm2_square"),
         step_pattern = c("symmetric2", "symmetric1"), normalize = TRUE,
         ws = NULL, threshold = NULL,
         return_matrix = FALSE, ncores = NULL, useRcppParallel = TRUE)
dtw_disvec(Q, lot, dist_method = c("norm1", "norm2", "norm2_square"),
         step_pattern = c("symmetric2", "symmetric1"), normalize = TRUE,
         ws = NULL, threshold = NULL, ncores = NULL)

Value

input

the function input parameters

dismat

the matrix of pairwise DTW distances, either as matrix or dist object

disvec

the vector DTW distances

Arguments

Q

time series, vector (univariate) or matrix (multivariate)

lot

List of time series. Each entry of the list is a time series as described in dtw2vec.

dist_method

character, describes the method of distance measure. See also dtw.

step_pattern

character, describes the step pattern. See also dtw.

normalize

logical, whether to return normalized pairwise distances or not. If step_pattern == 'symmetric1' only non-normalized distances can be returned (default = TRUE)

ws

integer, describes the window size for the sakoe chiba window. If NULL, then no window is applied. (default = NULL)

threshold

numeric, the threshold for early abandoning. In the calculation of the global cost matrix a possible path stops as soon as the threshold is reached. Facilitates faster calculations in case of low threshold. (default = FALSE)

return_matrix

logical, If FALSE (default) the distance matrix is returned as dist object. If TRUE a symmetric matrix of differences is returned.

ncores

integer, number of cores to be used for parallel computation of the distance matrix. If ncores = NULL (default) then ncores is set to the number of available cores minus 1. If ncores = 0 then no parallel computation is performed and standard sapply instead of parallel::parSapply is applied.

useRcppParallel

logical, if the package RcppParallel (TRUE, default) or parallel (FALSE) is used for parallel computation

Details

By setting the parameter return_matrix = FALSE (default) the output value dismat of dtw_dismat is a dist object and can easily be passed to standard clustering functions (see examples).

No matrices are allocated for calculating the pairwise distances.

Examples

if (FALSE) {

#--- Example for clustering a set of time series by feeding well known 
# clustering methods with DTW-distance objects. First we simulate 
# two prototype random walks and some cluster members. The cluster
# members are simulated by adding noise and randomly stretching and 
# comressing the time series, to get time warped time series of 
# varying lengths. The built clusters are 1:6 and 7:12.
set.seed(123)
N <- 100
rw_a <- cumsum(rnorm(N))
rw_b <- cumsum(rnorm(N))
sth <- sample(seq(0, 0.2, 0.01), size = 10)
cmp <- sample(seq(0, 0.2, 0.01), size = 10)
lot <- c(list(rw_a), 
         lapply(1:5, function(i){ 
           simulate_timewarp(rw_a + rnorm(N), sth[i], cmp[i]) 
         }),
         list(rw_b),
         lapply(6:10, function(i){ 
           simulate_timewarp(rw_b + rnorm(N), sth[i], cmp[i]) 
         }))


# Next get the distance matrix, as dist object. Per default all 
# minus 1 available cores are used:
result <- dtw_dismat(lot = lot, dist_method = "norm2", ws = 50, 
                     return_matrix = FALSE)
class(result$dismat)


# Finally you can cluster the result with the following
#  well known methods:
require(cluster)
myclus <- hclust(result$dismat)
plot(myclus)
summary(myclus)

myclus <- agnes(result$dismat)
plot(myclus)
summary(myclus)

myclus <- pam(result$dismat, k=2)
plot(myclus)
summary(myclus)
myclus$medoids


}