0

I would like to change the default step-pattern weight of the cost function because I need to standardize my results with some others in a paper that doesn't use the weight 2 for the diagonal distance. I've read the JSS paper but I just found other step-patterns that are not what I'm really looking for, I guess. For example, imagine we have two timeSeries Q, C:

Q = array(c(0,1,0,0,0,0,0,0,0,0,0,0,0,1,1,0),dim=c(8,2))
C = array(c(0,1,0,0,0,0,0,0,0,1,1,0,0,0,0,0),dim=c(8,2))

When I calculate the dtw distance, I obtain alignment = dtw(Q,C,keep=TRUE) With a alginment$distance of 2.41 and a cost matrix where for example the [2,2] element is 2 instead of 1 because of the weight or penalization of 2*d[i,j] in the diagonal when selecting the minimum between:

g[i,j] = min( g[i-1,j-1] + 2 * d[i  ,j  ] ,
              g[i  ,j-1] +     d[i  ,j  ] ,
              g[i-1,j  ] +     d[i  ,j  ] )
gissemari
  • 11
  • 2
  • You should provide a minimal [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example) with sample data showing the functions you are using in order to make it easier to help you. – MrFlick Sep 25 '16 at 22:22

1 Answers1

1
plot(asymmetricP1)
edit(asymmetricP1)
structure(c(1, 1, 1, 2, 2, 3, 3, 3, 1, 0, 0, 1, 0, 2, 1, 0, 2, 
1, 0, 1, 0, 1, 0, 0, -1, 0.5, 0.5, -1, 1, -1, 1, 1), .Dim = c(8L, 4L), class = "stepPattern", npat = 3, norm = "N")

Look at the plot, and consider the branches as ordered from right to left (ie. branch1 = 0.5 weight) Everything in the script below is in the context of plot(asymmetricP1) and edit(asymmetricP1) 

#first 8 digit sequence (1,1,1,2,2,3,3,3....
#branch1: "1,1,1" <- amount of intervals assigned to specificaly branch1; (end, joint, origin) 
#branch2: "2,2" <- only 2 intervals, this is the middle diagnol line.
#branch3: "3,3,3" <- amount of interals
#note: Don't be confused by the numbers themselves, ie. "4,4,4" <- 3 intervals; "2,2,2" <- 3 intervals

#for the next sequences consider:
#the sequence of each branch is to be read as farthest from origin -> 0,0 
#each interval assignment is accounted for in this order

#next 8 digit sequence: 1, 0, 0, 1, 0, 2, 1, 0,
#branch1: 1,0,0 <- interval position in relation to the query index
#branch2: 1,0 <- interval position in relation to the query index
#branch3: 2,1,0 <- interval position in relation to the query index (again see in plot)

#next 8 digit sequence: 2, 1, 0, 1, 0, 1, 0, 0
#branch1: 2,1,0 <- interval position in relation to the REFERENCE index
#branch2: 1,0 <- interval position in relation to the reference index
#branch3: 1,0,0 <- interval position in relation to the reference index (again see in plot)

#next 8 digit sequence: -1, 0.5, 0.5, -1, 1, -1, 1, 1
#note: "-1" is a signal that indicates weighting values follow
#note: notice that for each -1 that occurs, there is one value less, for example branch 1
# .....which has 3 intervals can only contain 2 weights (0.5 and 0.5)
#branch1: -1,0.5,0.5 <- changing the first 0.5 changes weight of [-1:0] segment (query index)
#branch2: -1,1 <- weight of middle branch
#branch3: -1,1,1 <- changing the second 1 changes weight of[-1,0] segment (query index)

#.Dim=c(8L, 4L): 
#8 represents the number of intervals (1,1,1,2,2,3,3,3)
#4 (from what I understand) is the (length of all the branch sequences mentioned previously)/8

#npat = 3
#3 is the number of patterns you described in the structure. ie(1,1,1,2,2,3,3,3)

Hope this helps, good luck!

Aaron43
  • 11
  • 2