Question

I am trying to define a function for ‘stickiness’ – a Business Analytics metric that measures user Engagement – and my function is returning a dataframe that is populated with unexpected data.

stickiness <- function(tdata) {
    require(plyr)
    mau_unique <- dlply(.data = tdata,
                        .variables = "dt",
                        .fun = function(x){unique(x$username)})
    dates_char <- names(mau_unique)
    dates_vector <- as.Date(dates_char[28:(length(dates_char))],
                            format = "%Y-%m-%d")
    output_df <- data.frame(dates_vector,
                            matrix(data = 0,
                                   nrow = length(dates_char) - 27,
                                   ncol = 3))
    colnames(output_df) <- c("Date", "DAU", "MAU", "Stickiness")
    for (i in 1:length(dates_vector)) {
        dt <- dates_vector[i]
        output_df[i, "DAU"] <- length(unlist(mau_unique[[as.character(dt)]][2]))
        set28 <- unique(unlist(lapply(X = mau_unique[i:(i + 27)], FUN = "[[", 2)))  
        output_df[i, "MAU"] <- length(set28)
        output_df[i, "Stickiness"] <- output_df[i, "DAU"] / output_df[i, "MAU"]
    }
    return(output_df) 
}

The following is returned:

         Date DAU MAU Stickiness
1  2012-04-28   1  28 0.03571429
2  2012-04-29   1  28 0.03571429
3  2012-04-30   1  28 0.03571429
4  2012-05-01   1  28 0.03571429
5  2012-05-02   1  28 0.03571429
6  2012-05-03   1  28 0.03571429
7  2012-05-04   1  28 0.03571429
8  2012-05-05   1  28 0.03571429
9  2012-05-06   1  28 0.03571429
10 2012-05-07   1  28 0.03571429

I expected something like the following:

         Date   DAU    MAU Stickiness
1  2012-04-28 25000 250000 0.10000000
...  ...      ...   ...    ...
10 2012-05-07 27371 284114 0.09633809

I suspect that the problem is related to which environments I’m evaluating in.

UPDATED sample data:

> tdata
                 dt  username
    4236 2012-04-06 241343664
    3091 2012-04-06 306001012
    2936 2012-04-06 388682041
    5790 2012-04-05 235612064
    6763 2012-04-05  69650072
    3392 2012-04-06    617142
    7684 2012-04-05 189752749
    3904 2012-04-06 255852653
    7915 2012-04-05 182713266
    6107 2012-04-05 187675644

UPDATE working function (using Brian Diggs’s answer):

stickiness <- function(tdata) {
    require(plyr)
    mau_unique <- dlply(.data = tdata,
                        .variables = "dt",
                        .fun = function(x){unique(x$username)})
    dates_char <- names(mau_unique)
    dates_vector <- as.Date(dates_char[28:(length(dates_char))],
                            format = "%Y-%m-%d")
    output_df <- data.frame(dates_vector,
                            matrix(data = 0,
                                   nrow = length(dates_char) - 27,
                                   ncol = 3))
    colnames(output_df) <- c("Date", "DAU", "MAU", "Stickiness")
    for (i in 1:length(dates_vector)) {
        dt <- dates_vector[i]
        output_df[i, "DAU"] <- length((mau_unique[[as.character(dt)]])
        set28 <- unique(do.call(c, mau_unique[i:(i + 27)]))  
        output_df[i, "MAU"] <- length(set28)
        output_df[i, "Stickiness"] <- output_df[i, "DAU"] / output_df[i, "MAU"]
    }
    return(output_df) 
}

Answer 1

Thanks for adding some sample data, but it is still not really reproducible since the function assumes the data spans at least 28 days (or rather, at least 28 unique dates).

The problem, as near as I can figure, is inside your for loop. With your example data,

> mau_unique
$`2012-04-05`
[1] 235612064  69650072 189752749 182713266 187675644

$`2012-04-06`
[1] 241343664 306001012 388682041    617142 255852653

attr(,"split_type")
[1] "data.frame"
attr(,"split_labels")
          dt
1 2012-04-05
2 2012-04-06

so in computing DAU, you pull a corresponding element from mau_unique. Working outward through your calculation of DAU with a dummy value for dt:

> dt <- as.Date("2012-04-05")
> dt
[1] "2012-04-05"
> as.character(dt)
[1] "2012-04-05"
> mau_unique[[as.character(dt)]]
[1] 235612064  69650072 189752749 182713266 187675644
> mau_unique[[as.character(dt)]][2]
[1] 69650072
> unlist(mau_unique[[as.character(dt)]][2])
[1] 69650072
> length(unlist(mau_unique[[as.character(dt)]][2]))
[1] 1

I don’t know how DAU should be calculated, but you always take the second username from the corresponding vector in mau_unique and take the length of that, which is why you always get 1. You are doing something similar for set28; I don’t know why you keep trying to pull the second element out.

---

EDIT:

Synthetically generated data is fine. That is a good way to create a lot of data in a small space, and with setting a random seed will allow everyone to work with the same data.

set.seed(1234)
tdata <- data.frame(dt = sample(seq(as.Date("2012-04-01"),
                                    as.Date("2012-04-30"),
                                    by = "day"),
                                size = 10000,
                                replace = TRUE),
                    username = sample(10000:10200,
                                      10000,
                                      replace = TRUE))

Given you descriptions of DAU and MAU, I think your for loop should read: (the rest of the function is unchanged)

for (i in 1:length(dates_vector)) {
    dt <- dates_vector[i]
    output_df[i, "DAU"] <- length(mau_unique[[as.character(dt)]])
    output_df[i, "MAU"] <- length(unique(unlist(mau_unique[i:(i+27)])))
    output_df[i, "Stickiness"] <- output_df[i, "DAU"] / output_df[i, "MAU"]
}

given this, your stickiness is:

> stickiness(tdata)
        Date DAU MAU Stickiness
1 2012-04-28 156 201  0.7761194
2 2012-04-29 168 201  0.8358209
3 2012-04-30 152 201  0.7562189