Create group variable based on common dates - r

Question

I have a large data set containing animal ID's and dates. There are two groups within this dataset but there is no grouping variable, so I have to extrapolate who belongs to which group based on the dates they appear to have in common.
Dummy data.
mydf<-data.frame(
Date=sort(rep(seq(as.Date("2012/1/1"),as.Date("2012/1/4"), length.out = 4),5)),
ID = c(1,2,3,4,5,5,6,7,8,9,1,2,3,4,5,6,7,8,9,10))
The other issue I have is that every now and then an ID belonging to group 1 might appear with a date associated with group 2, which is what has thrown off every attempt I've made so far at grouping.
What I need is a output with ID's and a new Group ID like this
ID Group
1 1
2 1
3 1
4 1
5 1
6 2
7 2
8 2
9 2
10 2
1:5 all appear together on the 1st and the 3rd so they are likely to be one group.
6:10 appear on the 2nd and 4th and are likely to be the 2nd group.
ID 5 belongs to group 1, because even though it was observed once on the 2nd with ID's 6:9, it was observed twice on the 1st and 2nd 1:4, so it's most likely to belong to group 1.
All my attempts have fallen flat. Can anyone offer a solution to this?
Thanks in advance.
EDIT:
I thought we had nailed a solution using Jon's kmeans solution (in the comments below):
mydf_wide <- mydf %>%
select(ID, date) %>%
distinct(ID,date)%>% #
mutate(x = 1) %>%
spread(date, x, fill = 0)
mydf_wide$clusters <- mydf_wide %>%
kmeans(centers = 2) %>%
pluck("cluster")
but I'm actually finding the kmeans method not quite getting it right every time. See below:
The groups where certain tags (ID) appear on the same day as each other are fairly easy to spot by eye. There are two groups, one is in the center, and the other group appears on either side. The clustering should be vertical by common dates as in Jon's answer below, but it is clustering across the entire date range. (Apologies for the messy axis labels)
The k-means method has worked on other groups, but it's not consistently able to group by common dates. I think the clustering approach is sensible, but I was wondering if there may be other clustering methods that may cope better than kmeans?
Alternatively, could a filtering method help reduce any background noise and help the kmeans approach more reliable?
Again, very grateful for any and all advice.
Cheers.

Answer 1

My thinking here is that you just assign each Date to a group, then take the average of group for each ID. You could then round to the nearest whole number from there. In this case, average group of ID == 5 would be 1.33
library(dplyr)
mydf %>%
mutate(group = case_when(
Date %in% as.Date(c("2012-01-01", "2012-01-03")) ~ 1,
Date %in% as.Date(c("2012-01-02", "2012-01-04")) ~ 2,
TRUE ~ NA_real_
)) %>%
group_by(ID) %>%
summarise(likely_group = mean(group) %>% round)
Which gives you the following:
# A tibble: 10 x 2
ID likely_group
<dbl> <dbl>
1 1 1
2 2 1
3 3 1
4 4 1
5 5 1
6 6 2
7 7 2
8 8 2
9 9 2
10 10 2
This works as long as there isn't an even split between groups for a single ID. But there isn't currently a way to address this situation with the information provided.

Answer 2

As a general solution, you might consider using k-means as an automatic way to split the data into groups based on similarity to other IDs.
First, I converted the data into wide format so that each ID gets one row. Then fed that into the base kmeans function to get the clustering output as a list, and purrr::pluck to extract just the assignment part of that list.
library(tidyverse)
mydf_wide <- mydf %>%
mutate(x = 1) %>%
spread(Date, x, fill = 0)
mydf_wide
# ID 2012-01-01 2012-01-02 2012-01-03 2012-01-04
#1 1 1 0 1 0
#2 2 1 0 1 0
#3 3 1 0 1 0
#4 4 1 0 1 0
#5 5 1 1 1 0
#6 6 0 1 0 1
#7 7 0 1 0 1
#8 8 0 1 0 1
#9 9 0 1 0 1
#10 10 0 0 0 1
clusters <- mydf_wide %>%
kmeans(centers = 2) %>%
pluck("cluster")
clusters
# [1] 2 2 2 2 2 1 1 1 1 1
Here's what that looks if you add those to the original data and plot.
mydf_wide %>%
mutate(cluster = clusters) %>%
# ggplot works better with long (tidy) data...
gather(date, val, -ID, -cluster) %>%
filter(val != 0) %>%
arrange(cluster) %>%
ggplot(aes(date, ID, color = as.factor(cluster))) +
geom_point(size = 5) +
scale_y_continuous(breaks = 1:10, minor_breaks = NULL) +
scale_color_discrete(name = "cluster")