Converting a raw data frame into workable time series - r

Question

I have a spreadsheet documenting prices of 40 similar products at various dates. It looks like this.
date_1<-seq(as.Date("2010-01-01"), as.Date("2011-01-01"), length.out = 40)
date_2<-seq(as.Date("2011-01-01"), as.Date("2012-01-01"), length.out = 40)
date_3<-seq(as.Date("2012-01-01"), as.Date("2013-01-01"), length.out = 40)
date_4<-seq(as.Date("2013-01-01"), as.Date("2014-01-01"), length.out = 40)
date_5<-seq(as.Date("2014-01-01"), as.Date("2015-01-01"), length.out = 40)
date_6<-seq(as.Date("2015-01-01"), as.Date("2016-01-01"), length.out = 40)
price_1<-floor(seq(20, 50, length.out = 40))
price_2<-floor(seq(20, 60, length.out = 40))
price_3<-floor(seq(20, 70, length.out = 40))
price_4<-floor(seq(30, 80, length.out = 40))
price_5<-floor(seq(40, 100, length.out = 40))
price_6<-floor(seq(50, 130, length.out = 40))
data.frame(date_1,price_1,date_2,price_2,date_3,price_3,date_4,price_4,date_5,price_5,date_6,price_6)
The problem is, the columns representing dates and prices alternate (convenient for record keeping). How can I transform the above data to a new dataframe consisting solely of prices of these 40 products as rows, with dates as column names? This will generate a lot of NA's because the dates in each column differ but that's OK.

Answer 1

When working with time series data it is often helpful to have it in long form (one row per observation), even if your target output is wide (one row per time series). Here are three possible approaches to get it into long form, then widen:
1. base reshape()
To get long form, base reshape is definitely a powerful option. The following solution improves on the accepted solution because it works for any numbers of products and observations and eliminates an unnecessary step:
df <- data.frame(date_1,price_1,date_2,price_2,date_3,price_3,
date_4,price_4,date_5,price_5,date_6,price_6)
# no need to create an id variable
long_form <- reshape(df, # idvar="id" by default
varying = list(grep('date_',names(df), value=TRUE),
grep('price_',names(df), value=TRUE) ),
v.names=c("date","price"),
direction="long",
sep="_")
And reshape can also widen it. (We'll use spread in another approach below.)
wide_form <- reshape(long_form, drop='time', timevar='date', direction='wide')
2. data.table melt() and dcast() (likely faster on real dataset)
Make sure you have data.table v1.9.6 or later, which allows you to melt multiple columns.
library(data.table)
setDT(df)
melt.data.table(df[, prod_id := .I], # product id = original row number
measure.vars = list(grep('date_',names(df), value=TRUE),
grep('price_',names(df), value=TRUE) ),
variable.name = 'sequence',
value.name = c('date','price'),
id.vars = 'prod_id') -> long_form
In this case you don't use the sequence, so to get wide form is just:
dcast.data.table(long_form[, !'sequence', with=FALSE],
value.var = 'price', # optional (function guesses correctly)
prod_id ~ date) -> wide_form
3. tidyr & dplyr split-apply-combine (easy to understand)
It doesn't require the mental gymnastics that reshape does (at least for me). It is a column-wise variant on the "split-apply-combine" paradigm.
library(dplyr); library(tidyr)
# Create long-form time series data
# Split table into sequenced prices and dates, then combine on product and sequence
full_join(
df %>%
select(starts_with('date_')) %>% #~~~~ Left side = date component ~~~~~~~~
mutate(prod_id = 1:nrow(df)) %>% #~ product id = original row number ~
gather(sequence, date, -prod_id) %>% #~ long form = 1 row per prod per seq ~
mutate(sequence = #~~~ Cols: product_id, sequence, date ~~~
sub('^date_(\\d+)$', '\\1', sequence) ) ,
df %>%
select(starts_with('price_')) %>% #~~~ Right side = price component ~~~~~~~
mutate(prod_id = 1:nrow(df)) %>% #~ ~
gather(sequence, price, -prod_id) %>% #~ same idea ~
mutate(sequence = #~~ Cols: product_id, sequence, price ~~~
sub('^price_(\\d+)$', '\\1', sequence) )
) -> long_form
In this case you don't need the sequence, so to get to wide form it's simply:
long_form %>% select(-sequence) %>% spread(date, price) -> wide_form
as noted by others above.

Answer 2

Here is one way I came up with using dplyr/tidyr packages:
library(tidyr)
library(dplyr)
date_1<-seq(as.Date("2010-01-01"), as.Date("2011-01-01"), length.out = 40)
date_2<-seq(as.Date("2011-01-01"), as.Date("2012-01-01"), length.out = 40)
date_3<-seq(as.Date("2012-01-01"), as.Date("2013-01-01"), length.out = 40)
date_4<-seq(as.Date("2013-01-01"), as.Date("2014-01-01"), length.out = 40)
date_5<-seq(as.Date("2014-01-01"), as.Date("2015-01-01"), length.out = 40)
date_6<-seq(as.Date("2015-01-01"), as.Date("2016-01-01"), length.out = 40)
price_1<-floor(seq(20, 50, length.out = 40))
price_2<-floor(seq(20, 60, length.out = 40))
price_3<-floor(seq(20, 70, length.out = 40))
price_4<-floor(seq(30, 80, length.out = 40))
price_5<-floor(seq(40, 100, length.out = 40))
price_6<-floor(seq(50, 130, length.out = 40))
df <- data.frame(date_1,price_1,date_2,price_2,date_3,price_3,date_4,price_4,date_5,price_5,date_6,price_6)
dates <- df[, grep('date', names(df))]
dates <- dates %>% gather(date_type, date) %>% select(-date_type)
prices <- df[, grep('price', names(df))]
prices <- prices %>% gather(price_type, price) %>% select(-price_type)
df <- cbind(dates, prices)
Then, to spread dates to columns and prices to rows, you can do something like this:
df <- arrange(df, price)
df <- spread(df, date, price)

Answer 3

Using baseR and tidyr you could do:
library(tidyr)
#add an id to identify the products
df$id=1:40
#transform the data to a long format
long_data <- reshape(df,idvar="id",varying=list(paste0("date_",1:6),paste0("price_",1:6)),v.names=c("date","price"),direction="long",sep="_")
long_data <- long_data[,!grepl("time",colnames(long_data))]
#put it back to a wide format
wide_data <- spread(long_data,date,price)