CODEWITHSUNDEEP
rtime-series
user1614738
user1614738

Reputation: 41

The hourly mean in time series

this is a time series with hourly smart meter data and freq=24. It is measured over three days, so first day[1:24], second[25:48], third[49:72].

I want to have the mean for every hour over three days. For example:

(t[1]+t[25]+t[49])/3

so I can make a boxplot for 24 mean hours over 3 days. 

x <- c(0.253, 0.132, 0.144, 0.272, 0.192, 0.132, 0.209, 0.255, 0.131, 
  0.136, 0.267, 0.166, 0.139, 0.238, 0.236, 1.75, 0.32, 0.687, 
  0.528, 1.198, 1.961, 1.171, 0.498, 1.28, 2.267, 2.605, 2.776, 
  4.359, 3.062, 2.264, 1.212, 1.809, 2.536, 2.48, 0.531, 0.515, 
  0.61, 0.867, 0.804, 2.282, 3.016, 0.998, 2.332, 0.612, 0.785, 
  1.292, 2.057, 0.396, 0.455, 0.283, 0.131, 0.147, 0.272, 0.198, 
  0.13, 0.19, 0.257, 0.149, 0.134, 0.251, 0.215, 0.133, 1.755, 
  1.855, 1.938, 1.471, 0.528, 0.842, 0.223, 0.256, 0.239, 0.113)

Upvotes: 3

Views: 3079

Answers (2)

Paul Hiemstra
Paul Hiemstra

Reputation: 60984

Because you did not post an easy to use set of example data, let's first generate some:

time_series = runif(72)

The next step would be to change the structure of the dataset from a 1d vector, to a 2d matrix, this saves you a lot of having to deal with indices and such:

time_matrix = matrix(time_series, 24, 3)

and use apply to calculate the hourly means (if you like apply, take a look at the plyr package for more nice functions, see this link for more detail):

hourly_means = apply(time_matrix, 1, mean)
> hourly_means
 [1] 0.2954238 0.6791355 0.6113670 0.5775792 0.3614329 0.4414882 0.6206761
 [8] 0.2079882 0.6238492 0.4069143 0.6333607 0.5254185 0.6685191 0.3629751
[15] 0.3715500 0.2637383 0.2730713 0.3170541 0.6053016 0.6550780 0.4031117
[22] 0.6857810 0.4492246 0.4795785

However, if you use ggplot2 there is no need to precalculate the boxplots, ggplot2 does this for you:

require(ggplot2)
require(reshape2)
# Notice the use of melt to reshape the dataset a bit
# Also notice the factor to transform Var1 to a categorical dataset
ggplot(aes(x = factor(Var1), y = value), 
       data = melt(time_matrix)) + 
       geom_boxplot()

Which yields, what I think, you where after:

enter image description here

On the x-axis the hours of the day, on the y axis the value.

Note: the data you have is a timeseries. R has specific ways of dealing with timeseries, e.g. the ts function. I normally use ordinary R data objects (array's, matrices), but you could take a look at the TimeSeries CRAN taskview for an overview of what R can do with timeseries.

To calculate the hourly means using a ts object (inspired by this SO post):

# Create a ts object
time_ts = ts(time_series, frequency = 24)
# Calculate the mean
> tapply(time_ts, cycle(time_ts), mean)
        1         2         3         4         5         6         7         8 
0.2954238 0.6791355 0.6113670 0.5775792 0.3614329 0.4414882 0.6206761 0.2079882 
        9        10        11        12        13        14        15        16 
0.6238492 0.4069143 0.6333607 0.5254185 0.6685191 0.3629751 0.3715500 0.2637383 
       17        18        19        20        21        22        23        24 
0.2730713 0.3170541 0.6053016 0.6550780 0.4031117 0.6857810 0.4492246 0.4795785 
> aggregate(as.numeric(time_ts), list(hour = cycle(time_ts)), mean)
   hour         x
1     1 0.2954238
2     2 0.6791355
3     3 0.6113670
4     4 0.5775792
....

Upvotes: 10

Joshua Ulrich
Joshua Ulrich

Reputation: 176728

You can do this easily with the boxplot function that comes with a basic R installation. Just create a data.frame with your original series and an index to identify the hour of each day.

Data <- data.frame(series=x, time=rep(1:24,3))
boxplot(series ~ time, data=Data)

boxplot output

Upvotes: 3

Related Questions