Creating time intervals using sqldf package

Question 1

# generate example data, 30 min intervals
set.seed(1)
dateseq <- seq(as.POSIXct("2013-11-17"), as.POSIXct("2013-11-18"), by="min")
df <- data.frame(Time=dateseq[sample(1:length(dateseq), 500)],
                 No_Words=sample(1:100, 500, replace=T))
groups <- cut.POSIXt(df$Time, breaks="30 min")

The hard way using sqldf:

library(sqldf)
df$groups <- groups
agg <- sqldf("select groups, avg(No_Words) from df group by groups", row.names=T)
row.names(agg) <- agg[,1]
agg <- as.matrix(agg)
class(agg) <- "numeric"
par(mar=c(2,10,0,0)); barplot(agg[,2], horiz=TRUE, las=1)

The easy way using e.g. tapply:

agg <- tapply(df$No_Words, list(groups), mean)
par(mar=c(2,10,0,0)); barplot(agg, horiz=TRUE, las=1)

Question 2

sqldf Here is an sqldf solution where the input data frame is DF:

library(sqldf)

min15 <- 15 * 60 # in seconds
ans <- fn$sqldf("select
       t.Time - t.Time % $min15 as Time, 
       sum(t.No_Words) as No_Words
    from DF t 
    group by Time")
plot(No_Words ~ Time, ans, type = "o")

giving:

> ans
                 Time No_Words
1 2013-11-17 13:30:00       52
2 2013-11-17 14:00:00       29
3 2013-11-17 17:15:00       39
4 2013-11-17 22:45:00       31

With dense grid If a dense grid is wanted then we will need a grid data frame, G, which we join with the prior ans (Note that sqldf pulls in the chron package so we use its trunc function):

# create grid G
rng <- range(as.POSIXct(trunc(as.chron(DF$Time), 15 / (24 * 60))))
G <- data.frame(Time = seq(rng[1], rng[2], by = min15))

ans2 <- sqldf("select Time, coalesce(No_Words, 0) as No_Words 
         from (select * from G left join ans using(Time))")
plot(No_Words ~ Time, ans2, type = "o")

The first few rows of ans2 are:

> head(ans2)

                 Time No_Words
1 2013-11-17 13:30:00       52
2 2013-11-17 13:45:00        0
3 2013-11-17 14:00:00       29
4 2013-11-17 14:15:00        0
5 2013-11-17 14:30:00        0
6 2013-11-17 14:45:00        0

zoo We also show a zoo solution:

library(zoo)
library(chron)
FUN <- function(x) as.POSIXct(trunc(as.chron(x), 15 / (24 * 60)))
z <- read.zoo(DF, FUN = FUN, aggregate = sum)
plot(z)

which gives for z:

> z
2013-11-17 13:30:00 2013-11-17 14:00:00 2013-11-17 17:15:00 2013-11-17 22:45:00 
             52                  29                  39                  31

Note: We used this data and, in particular, Time is of class "POSIXct":

Lines<- " Time            No_Words
1   2013-11-17 13:37:00                    6    
2   2013-11-17 13:37:00                   16    
3   2013-11-17 13:37:00                   18    
4   2013-11-17 13:37:00                   12    
5   2013-11-17 14:03:00                    5    
6   2013-11-17 14:03:00                   20    
7   2013-11-17 14:04:00                    4    
8   2013-11-17 17:21:00                   39    
9   2013-11-17 22:48:00                   19    
10  2013-11-17 22:48:00                   12   
"

raw <- read.table(text = Lines, skip = 1)
DF <- data.frame(Time = as.POSIXct(paste(raw$V2, raw$V3)), No_Words = raw$V4)

Question 3

This answer is not with sqldf, but with the base R functions aggregate and cut:

## If your "Time" column is not an actual time object, 
##    convert it to one before proceeding.
mydf$Time <- as.POSIXct(mydf$Time)

cut can create time bins. We'll use that to do our aggregation. You could use the formula notation, but I've used the list approach so that it is easy to specify the resulting column names:

## Aggregate data in 30 minute chunks
aggregate(list(No_Words = mydf$No_Words), 
          list(Time = cut(mydf$Time, "30 min")), FUN = mean)
#                  Time No_Words
# 1 2013-11-17 13:37:00 11.57143
# 2 2013-11-17 17:07:00 39.00000
# 3 2013-11-17 22:37:00 15.50000

## Aggregate data into 15 minute chunks
aggregate(list(No_Words = mydf$No_Words), 
          list(Time = cut(mydf$Time, "15 min")), FUN = mean)
#                  Time  No_Words
# 1 2013-11-17 13:37:00 13.000000
# 2 2013-11-17 13:52:00  9.666667
# 3 2013-11-17 17:07:00 39.000000
# 4 2013-11-17 22:37:00 15.500000