Assign the cluster number for each observation for time series data with hclust in R

Admin 1
Viewed 237

I have a time series data with 4 variables over about 5 years. I would like to cluster the data using hclust method in R. I would like to cluster the observations. My code works. However, I would like to identify the specific cluster for each observation. That is, I would like to add the number of the cluster beside each observation. My code gave me an error. I understand the error. So is there any way can I achieve my point.

Here is my try:

library(TSclust)
library(cluster)    # clustering algorithms
library(tseries)
library(zoo)
library(dtw)
library(dtwclust)
library(dplyr)
##Load the data
data("EuStockMarkets")
##Save the data
dat <- EuStockMarkets

res <- lapply(split(as.zoo(EuStockMarkets), as.integer(time(EuStockMarkets))), as.ts)
## Re-define the data
datNew <- ts(rbind(res$`1995`,res$`1996`,res$`1997`, res$`1998`))
d <- dist(datNew, method = "DTW")
hc1 <- hclust(d, method = "average" )
sub_grp <- cutree(hc1, k = 4)
table(sub_grp)
datNew%>%
  mutate(cluster = sub_grp) %>%
  head

It returns me an error:

 Error in UseMethod("mutate_") : 
 no applicable method for 'mutate_' applied to an object of class "c('mts', 'ts', 'matrix')"

In addition: Warning message:
`mutate_()` is deprecated as of dplyr 0.7.0.
Please use `mutate()` instead.
See vignette('programming') for more help
hierarchical-clustering dplyr time-series r
1 Answers

I suppose the problem is that your datNew is not a data.frame. Looking at:

class(datNew)
[1] "mts"    "ts"     "matrix"

And this gives to you your error. If you put it as a data.frame:

library(dplyr)
data.frame(timeseries=as.matrix(datNew), date=time(datNew))%>%
           # use mutate() instead of mutate_()
           mutate(cluster = sub_grp) %>%
           head()

It should work, hoping it's the result you need.

  timeseries.DAX timeseries.SMI timeseries.CAC timeseries.FTSE date cluster
1        2110.77         2673.5         1956.0          3083.4    1       1
2        2097.34         2656.2         1927.8          3095.8    2       1
3        2074.68         2628.8         1894.2          3065.6    3       1
4        2097.51         2628.8         1881.2          3065.5    4       1
5        2079.19         2628.8         1881.2          3065.5    5       1
6        2068.92         2612.3         1885.9          3065.7    6       1

EDIT

In case is needed you can try this using hclust on after log and dif:

res <- lapply(split(as.zoo(EuStockMarkets), as.integer(time(EuStockMarkets))), as.ts)
datNew <- ts(rbind(res$`1995`,res$`1996`,res$`1997`, res$`1998`))

dat.log <- log(datNew)
dat.diff <- diff(dat.log)
Logreturns <- dat.diff

# using a different dist, due an error, the idea is the same
d <- dist(Logreturns)

hc1 <- hclust(d, method = "average" )
sub_grp <- cutree(hc1, k = 4)


data.frame(timeseries=as.matrix(Logreturns), date=time(Logreturns))%>%
  mutate(cluster = sub_grp) %>% 
  head()
user_64986500
Related