CODEWITHSUNDEEP
rpackage
Yang Yang
Yang Yang

Reputation: 902

R package `bfast` error: missing value where TRUE/FALSE needed

I am using the R package bfast to detect any change point in a time series, and come across the following error. Is there any suggestion? Thanks a lot.

library(bfast)
Mydata = Nile
bf1 <- bfast01(data = Mydata)

Error in if (is.nan(p0) || p0 < a2 || p0 > (1 - a2)) { : missing value where TRUE/FALSE needed

Upvotes: 0

Views: 413

Answers (1)

Maurits Evers
Maurits Evers

Reputation: 50718

It seems that bfast01 tries to guess the model from the other function arguments. I'm not too familiar with bfast but this seems to lead to the error on the Nile data.

If we explicitly specify the model,

fit <- bfast01(Nile, formula = response ~ trend);

there won't be an error, and bfast01 picks up a change point at index 28. 

fit$breakpoints;
#[1] 28

This result is consistent with results from a similar change point analysis using changepoint:

changepoint::cpt.mean(Nile, class = FALSE);
#    cpt conf.value
#     28          1

Note that for the bfast change point analysis, a model involving only a trend component seems sufficient, as the seasonal/autoregressive effects occur on a smaller scale. You'll need to check the validity of the model based on your real data.

Upvotes: 3

Related Questions