Subsetting observations with grouping some features - r

Question

I have a dataset like below:
date, time,product,shop_id
20140104 900 Banana 18
20140104 900 Banana 19
20140104 924 Banana 18
20140104 929 Banana 18
20140104 932 Banana 20
20140104 948 Banana 18
and i need to extract the observations with different product, and different shop_id
so, i need to group the observations by product+shop_id
here is my code:
library(plyr)
d_ply( shop, .( product,shop_id ),table )
print(p)
unfortunately, it prints null
dataset:
date=c(20140104,20140104,20140104,20140104,20140104)
time=c(924 ,900,854,700,1450)
product=c(Banana ,Banana ,Banana ,Banana ,Banana)
shop_id=c(18,18,18,19,20)
shop<-data.frame(date=date,time=time,product=product,shop_id=shop_id)
the output should be
date, time, product, shop_id
20140104 900 Banana 19
20140104 932 Banana 20
20140104 948 Banana 18

Answer 1

We can do
library(tidyverse)
shop %>%
group_by(product, shop_id) %>%
mutate(n = n()) %>%
group_by(time) %>%
arrange(n) %>%
slice(1) %>%
group_by(product, shop_id) %>%
arrange(-time) %>%
slice(1) %>%
select(-n) %>%
arrange(time)
# date time product shop_id
# <int> <int> <chr> <int>
#1 20140104 900 Banana 19
#2 20140104 932 Banana 20
#3 20140104 948 Banana 18

Answer 2

In order to take only first unique combination, just use aggregate from package stats:
> aggregate(shop, by=list(shop$product, shop$shop_id), FUN=function(x){x[1]})
Group.1 Group.2 date time product shop_id
1 Banana 18 20140104 924 Banana 18
2 Banana 19 20140104 700 Banana 19
3 Banana 20 20140104 1450 Banana 20
Explanation: My FUN=function(x){x[1]} takes only first element in case of collision
To drop "Group.1", "Group.2" or other columns:
> res <- aggregate(shop, by=list(shop$product, shop$shop_id), FUN=function(x){x[1]})
> res[ , !(names(res) %in% c("Group.1", "Group.2"))]
date time product shop_id
1 20140104 924 Banana 18
2 20140104 700 Banana 19
3 20140104 1450 Banana 20
P.S. Your dataset provided is inconsistent with examples you required, so that's why there is a difference in numbers.
P.S.2 If you want to get all data in case of collision:
> aggregate(shop, by=list(shop$product, shop$shop_id), FUN="identity")
Group.1 Group.2 date time product shop_id
1 Banana 18 20140104, 20140104, 20140104 924, 900, 854 1, 1, 1 18, 18, 18
2 Banana 19 20140104 700 1 19
3 Banana 20 20140104 1450 1 20
If you want to mark collisions:
> aggregate(shop, by=list(shop$product, shop$shop_id), FUN=function(x){if (length(x) > 1) NA else x})
Group.1 Group.2 date time product shop_id
1 Banana 18 NA NA NA NA
2 Banana 19 20140104 700 1 19
3 Banana 20 20140104 1450 1 20
If you want to exclude non-unique rows:
> res <- aggregate(shop, by=list(shop$product, shop$shop_id), FUN=function(x){if (length(x) > 1) NULL else x})
> res[res$product != "NULL", !(names(res) %in% c("Group.1", "Group.2"))]
date time product shop_id
2 20140104 700 1 19
3 20140104 1450 1 20
If you want to avoid coerce from String to Int (for product), use ""/"NULL"/"NA" instead of NULL/NA.

Answer 3

It can be done using dplyr as follows:
# create the sample dataset
date=c(20140104,20140104,20140104,20140104,20140104)
time=c(924 ,900,854,700,1450)
product=c("Banana","Banana","Banana","Banana","Banana")
shop_id=c(18,18,18,19,20)
shop<-data.frame(date=date,time=time,product=product,shop_id=shop_id)
# load a dplyr library
library(dplyr)
# take shop data
shop %>%
# group by product, shop id, date
group_by(product, shop_id, date) %>%
# for each such combination, find the earliest time
summarise(time = min(time)) %>%
# group by product, shop id
group_by(product, shop_id) %>%
# for each combination of product & shop id
# return the earliest date and time recorded on the earliest date
summarise(date = min(date), time = time[date == min(date)])