Mueen's Algorithm for Similarity Search is The Fastest Similarity Search Algorithm for Time Series Subsequences under Euclidean Distance and Correlation Coefficient.

dist_profile(
  data,
  query,
  ...,
  window_size = NULL,
  method = "v3",
  index = 1,
  k = NULL,
  weight = NULL,
  paa = 1
)

Arguments

data

a matrix or a vector.

query

a matrix or a vector. See details.

...

Precomputed values from the first iteration. If not supplied, these values will be computed.

window_size

an int or NULL. Sliding window size. See details.

method

method that will be used to calculate the distance profile. See details.

index

an int. Index of query window. See details.

k

an int or NULL. Default is NULL. Defines the size of batch for MASS V3. Prefer to use a power of 2. If NULL, it will be set automatically.

weight

a vector of numeric or NULL with the same length of the window_size. This is a MASS extension to weight the query.

paa

a numeric. Default is 1. Factor of PAA reduction (2 == half of size). This is a MASS extension.

Value

Returns the distance_profile for the given query and the last_product for STOMP algorithm and the parameters for recursive call. See details.

Details

This function has several ways to work:

Case 1: You have a small sized query and the data. In this case you only have to provide the first two parameters data and query. Internally the window_size will be get from the query length.

Case 2: You have one or two data vectors and want to compute the join or self-similarity. In this case you need to use the recursive solution. The parameters are data, query, window_size and index. The first iteration don't need the index unless you are starting somewhere else. The query will be the source of a query_window, starting on index, with length of window_size.

The method defines which MASS will be used. Current supported values are: v2, v3, weighted.

References

  • Abdullah Mueen, Yan Zhu, Michael Yeh, Kaveh Kamgar, Krishnamurthy Viswanathan, Chetan Kumar Gupta and Eamonn Keogh (2015), The Fastest Similarity Search Algorithm for Time Series Subsequences under Euclidean Distance

Website: https://www.cs.unm.edu/~mueen/FastestSimilaritySearch.html

Examples


w <- mp_toy_data$sub_len
ref_data <- mp_toy_data$data[, 1]
# minimum example, data and query
nn <- dist_profile(ref_data, ref_data[1:w])
#> tsmp Parsing data 
#> Error in dist_profile(ref_data, ref_data[1:w]): Assertion on 'window_size' failed. Must be of class 'integerish', not 'double'.
distance_profile <- sqrt(nn$distance_profile)
#> Error in eval(expr, envir, enclos): object 'nn' not found

# data and indexed query
nn <- dist_profile(ref_data, ref_data, window_size = w, index = 10)
#> tsmp Parsing data +29ms 
#> tsmp Parsing type +2ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +1ms 
#> tsmp params is null +1ms 
#> tsmp params is a list with par and window_size +1ms 
distance_profile <- sqrt(nn$distance_profile)

# recursive
nn <- NULL

for (i in seq_len(10)) {
  nn <- dist_profile(ref_data, ref_data, nn, window_size = w, index = i)
}
#> tsmp Parsing data +6ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +0ms 
#> tsmp params is null +0ms 
#> tsmp params is a list with par and window_size +1ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +0ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +0ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +0ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +0ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +1ms 
#> tsmp params is not null +1ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +6ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +1ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +0ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +0ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +0ms 
#> tsmp params is not null +1ms 
#> tsmp Parsing data +2ms 
#> tsmp Parsing type +1ms 
#> tsmp Register anytime exit point +1ms 
#> tsmp Computation +1ms 
#> tsmp params is not null +0ms 

# weighted
weight <- c(rep(1, w / 3), rep(0.5, w / 3), rep(0.8, w / 3)) # just an example

nn <- dist_profile(ref_data, ref_data,
  window_size = w, index = 1, method = "weighted",
  weight = weight
)
#> Error in dist_profile(ref_data, ref_data, window_size = w, index = 1,     method = "weighted", weight = weight): unused argument (method = "weighted")
distance_profile <- sqrt(nn$distance_profile)