Remove dates which are not continuous in the data in R - r

Question

I have a dataframe and I want to filter out the entries that are not continuous in date. In other words, I am looking at the cluster of continuous dates.
a %>% group_by(day) %>% summarise(count = n()) %>% mutate(day_dif = day - lag(day))
Source: local data frame [20 x 3]
day count day_dif
(date) (int) (dfft)
1 2016-02-02 12 NA days
2 2016-02-03 80 1 days
3 2016-02-04 102 1 days
4 2016-02-05 97 1 days
5 2016-02-06 118 1 days
6 2016-02-07 115 1 days
7 2016-02-08 4 1 days
8 2016-02-20 13 12 days
9 2016-02-21 136 1 days
10 2016-02-22 114 1 days
11 2016-02-23 134 1 days
12 2016-02-24 126 1 days
13 2016-02-25 128 1 days
14 2016-02-26 63 1 days
15 2016-02-27 118 1 days
16 2016-03-06 1 8 days
17 2016-03-29 28 23 days
18 2016-04-03 18 5 days
19 2016-04-08 18 5 days
20 2016-04-27 23 19 days
In this, I want to filter out entries that are not continuous in date. For ex, 2016-03-06, 2016-03-29, 2016-04-03 are single day entries which needs to be removed. I am looking only for continuous days entries. entries that occur with multiple days. My ideal output which I am looking is,
day count day_dif Cluster
(date) (int) (dfft)
1 2016-02-02 12 NA days 1
2 2016-02-03 80 1 days 1
3 2016-02-04 102 1 days 1
4 2016-02-05 97 1 days 1
5 2016-02-06 118 1 days 1
6 2016-02-07 115 1 days 1
7 2016-02-08 4 1 days 1
8 2016-02-20 13 12 days 2
9 2016-02-21 136 1 days 2
10 2016-02-22 114 1 days 2
11 2016-02-23 134 1 days 2
12 2016-02-24 126 1 days 2
13 2016-02-25 128 1 days 2
14 2016-02-26 63 1 days 2
15 2016-02-27 118 1 days 2
Where cluster column indicates the date clusters and also the output removes the single dates. Here 1 in the cluster column indicates, first group of dates and 2 indicates second group of dates. If there are more than 3 continuous days, I want to consider as on cluster.
I am trying to do this by using lag functions and all. But without much success. Can anybody help me in doing this? Any idea would be appreciated.
Thanks

Answer 1

We can use rle to subset the rows
i1 <- c(TRUE, a1$day_dif[-1] >=3)
i2 <- inverse.rle(within.list(rle(i1), {values1 <- values
values[values1 &lengths >3] <- FALSE
values[!values1]<- TRUE}))
a1$Cluster <- cumsum(i1)
a1[i2,]
# day count day_dif Cluster
#1 2016-02-02 12 NA days 1
#2 2016-02-03 80 1 days 1
#3 2016-02-04 102 1 days 1
#4 2016-02-05 97 1 days 1
#5 2016-02-06 118 1 days 1
#6 2016-02-07 115 1 days 1
#7 2016-02-08 4 1 days 1
#8 2016-02-20 13 12 days 2
#9 2016-02-21 136 1 days 2
#10 2016-02-22 114 1 days 2
#11 2016-02-23 134 1 days 2
#12 2016-02-24 126 1 days 2
#13 2016-02-25 128 1 days 2
#14 2016-02-26 63 1 days 2
#15 2016-02-27 118 1 days 2
The above code can be also be chained (%>%)
a1 %>%
mutate(i1 = c(TRUE, day_dif[-1] >=3)) %>%
do(data.frame(., i2 = inverse.rle(within.list(rle(.$i1), {
values1 <- values
values[values1 & lengths >3] <- FALSE
values[!values1] <- TRUE
})))) %>%
mutate(Cluster = cumsum(i1)) %>%
filter(i2) %>%
select(-i1, -i2)
# day count day_dif Cluster
#1 2016-02-02 12 NA days 1
#2 2016-02-03 80 1 days 1
#3 2016-02-04 102 1 days 1
#4 2016-02-05 97 1 days 1
#5 2016-02-06 118 1 days 1
#6 2016-02-07 115 1 days 1
#7 2016-02-08 4 1 days 1
#8 2016-02-20 13 12 days 2
#9 2016-02-21 136 1 days 2
#10 2016-02-22 114 1 days 2
#11 2016-02-23 134 1 days 2
#12 2016-02-24 126 1 days 2
#13 2016-02-25 128 1 days 2
#14 2016-02-26 63 1 days 2
#15 2016-02-27 118 1 days 2
data
a <- structure(list(day = structure(c(16833, 16834, 16835, 16836,
16837, 16838, 16839, 16851, 16852, 16853, 16854, 16855, 16856,
16857, 16858, 16866, 16889, 16894, 16899, 16918), class = "Date"),
count = c(12L, 80L, 102L, 97L, 118L, 115L, 4L, 13L, 136L,
114L, 134L, 126L, 128L, 63L, 118L, 1L, 28L, 18L, 18L, 23L
)), .Names = c("day", "count"), row.names = c("1", "2", "3",
"4", "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15",
"16", "17", "18", "19", "20"), class = "data.frame")
a1 <- a %>%
mutate(day_dif = day - lag(day))

Answer 2

There is probably a better way to deal with the first NA values. Here, I manually assigned it to 0. Then, because the difference of continuous date will be 1, you can take advantage of this property to create a boolean vector and then use cumsum to get the results. Finally, you can remove those groups which their length are equal to 1.
# Let the first NA equal to 0
df[which(is.na(df), arr.ind=TRUE)] <- 0
df %>% mutate(cluster=cumsum(day_dif !=1)) %>%
group_by(cluster) %>% filter(length(cluster) > 1) %>% ungroup()
# Source: local data frame [15 x 4]
# day count day_dif cluster
# (date) (int) (dfft) (int)
# 1 2016-02-02 12 0 days 1
# 2 2016-02-03 80 1 days 1
# 3 2016-02-04 102 1 days 1
# 4 2016-02-05 97 1 days 1
# 5 2016-02-06 118 1 days 1
# 6 2016-02-07 115 1 days 1
# 7 2016-02-08 4 1 days 1
# 8 2016-02-20 13 12 days 2
# 9 2016-02-21 136 1 days 2
# 10 2016-02-22 114 1 days 2
# 11 2016-02-23 134 1 days 2
# 12 2016-02-24 126 1 days 2
# 13 2016-02-25 128 1 days 2
# 14 2016-02-26 63 1 days 2
# 15 2016-02-27 118 1 days 2