in order to cluster time series by certain criteria I want to measure the length of gaps inside of the series. These gaps may indicate various business problems and an experienced analyst should have a look at them, before the values are pipelined into the predictive model. I know how to identify, count and replace NA's. That's not what I want to do here. Rather I want to identify the length of each consecutive (the single ones too) NA value as a vector per time series.
Some code to illustrate my problem and wanted outcome:
suppressPackageStartupMessages(library(tidyverse))
suppressPackageStartupMessages(library(lubridate))
dates <- ymd("2016-01-01")+ months(0:71)
fake_values <-
c(661,678,1094,1987,3310,2105,1452,983,1107,805,675,684,436,514,668,206,19,NA,NA,NA,1174,1760,735,366,
510,580,939,1127,2397,1514,1370,832,NA,661,497,328,566,631,983,1876,2784,2928,2543,1508,1175,8,1733,
862,779,1112,1446,2407,3917,2681,NA,NA,NA,NA,NA,1239,1581,2814,1419,2792,4899,5857,2450,1944,1464,1248,
1533,NA)
df <- bind_cols(fake_values, dates) %>%
rename(c(values = ...1, dates = ...2)
)
# resulting vector('s)
result <- as.data.frame(c(3,1,5,1))
names(result)[1] = "some_products_gaps"
By the way, I would prefer a tidy solution. Hope I am not hoping too much... The solution must be scalable, as I need to loop/apply it for thousands of time series. Thanks in advance!
With the help of data.table rleid here's a dplyr solution.
library(dplyr)
df %>%
mutate(grp = data.table::rleid(is.na(values))) %>%
filter(is.na(values)) %>%
count(grp, name = 'some_products_gaps') %>%
select(-grp)
# some_products_gaps
# <int>
#1 3
#2 1
#3 5
#4 1