On Thu, 7 Mar 2019, Sparks, John wrote:
Thanks to Achim's direction I now have a re-producible example.
The code below creates a ts object. The x scale of the last graph runs
from 0 to 700.
Yes, and hence breakpoints() re-uses that scaling. As I wrote in my
previous mail you either have to squeeze your data into a regular grid of
52 weekly observations per year or you have to keep track of the time
index yourself. See below.
So just need a way to get that scale to show the weeks (or some summary
of them).
Thanks a bunch.
--JJS
library(strucchange)
library(xts)
library(lubridate)
#rm(list=ls())
data("Nile")
class(Nile)
plot(Nile)
bp.nile <- breakpoints(Nile ~ 1)
ci.nile <- confint(bp.nile, breaks = 1)
lines(ci.nile)
dfNile<-data.frame(as.numeric(Nile))
dfNile$week<-seq(ymd('2012-01-01'),ymd('2013-11-30'),by='weeks')
tsNile <- as.xts(x = dfNile[, -2], order.by = dfNile$week)
tsNile<-as.ts(tsNile)
plot(tsNile)
bp.tsNile <- breakpoints(tsNile ~ 1)
ci.tsNile <- confint(bp.tsNile, breaks = 1)
lines(ci.tsNile)
If you want to use your own non-ts time scale, you can use "xts" (as you
do above) or "zoo" (as I do below) or keep thing in a plain "data.frame"
(or similar). Then you just have to index the times with the breakpoints
or their confidence intervals respectively:
## zoo series
x <- zoo(Nile, seq(ymd('2012-01-01'),ymd('2013-11-30'),by='weeks'))
## breakpoints and confidence intervals
bp <- breakpoints(x ~ 1)
ci <- confint(bp, breaks = 1)
## map time index
cix <- time(x)[ci$confint]
## visualize
plot(x)
abline(v = cix[2], lty = 2)
arrows(cix[1], min(x), cix[3], min(x),
col = 2, angle = 90, length = 0.05, code = 3)
Above, the $confint is a vector. If it is a matrix (due to more than one
breakpoint) the code needs to be tweaked to make cix also a matrix and
then use cix[,i] rather than cix[i] for i = 1, 2, 3.
________________________________
From: Achim Zeileis <achim.zeil...@uibk.ac.at>
Sent: Wednesday, March 6, 2019 6:11 PM
To: Sparks, John
Cc: r-help@r-project.org
Subject: Re: [R] strucchange Graph By Week and xts error
On Thu, 7 Mar 2019, Sparks, John wrote:
Hi R Helpers,
I am doing some work at identifying change points in time series data.
A very nice example is given in the R Bloggers post
https://www.r-bloggers.com/a-look-at-strucchange-and-segmented/
The data for the aswan dam in that example is yearly. My data is
weekly. I ran the code switching the data for the analysis to my data
and it worked, but the scale of the line chart is not sensible. I have
225 weekly observations and the x-axis of the line graph shows numbers
from 0 to over 1500. The information on the ts object is
Unfortunately, breakpoints() can only deal automatically with "ts" time
series not with zoo/xts/... So either you can squeeze your data onto a
regular "ts" grid which may work in the case of weekly data. Or you need
to handle the time index "by hand". See
https://stackoverflow.com/questions/43243548/strucchange-not-reporting-breakdates/43267082#43267082
for an example for this.
As for the as.xts() error below. This is because dfNile[, -2] is still a
"ts" object and then as.xts() sets up "order.by" for you.
Either you use xts() rather than as.xts() or you make the first column in
the data.frame "numeric" rather than "ts", e.g., by starting the
transformation with:
dfNile<-data.frame(as.numeric(Nile))
Start=1
End=1569
Frequency=0.1428...
I can't share the data because it is proprietary.
Wanting to be a good member of the list, I attempted to put weekly
increments on the Nile data so I could reproduce the x axis of the chart
with the axis scale that I am seeing. Unfortunately, in doing so I got
another error that I don't understand.
library(strucchange)
library(lubridate)
library(xts)
# example from R-Blog runs fine
data(???Nile???)
plot(Nile)
bp.nile <- breakpoints(Nile ~ 1)
ci.nile <- confint(bp.nile, breaks = 1)
lines(ci.nile)
#problem comes in here
dfNile<-data.frame(Nile)
dfNile$week<-seq(ymd('2012-01-01'),ymd('2013-11-30'),by='weeks')
tsNile<-as.xts(x=dfNile[,-2],order.by=dfNile$week)
Error in xts(x.mat, order.by = order.by, frequency = frequency(x), ...) :
formal argument "order.by" matched by multiple actual arguments
Can somebody help me to put together the ts object with weeks so that I can
demonstrate the problem with the scale on the x-axis and then try to get some
help with that original problem?
Much appreciated.
--John Sparks
[[alternative HTML version deleted]]
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
