How to generate meaningful fake data for learning, experimentation and teaching

The Problem

There’s one thing about R that a lot of people have as their Top-of-Mind. That’s the black-and-white plot of iris dataset which is definitely a huge boring view of R. That’s boring because of aesthetics but also because it’s such a cliched example used over and over again. The other problem is finding the right set of dataset for the right set of problem you want to teach/learn/experiment. Let’s say you want to teach Time Series and that’s a case where your Spam / Ham Classification Dataset isn’t going to be of any use.

Solution

No more worries. That’s where fakir has arrived to help us. fakir is an R-package by Colin Fay (of Think-R) who’s been so good with his contributions to the R community.

About fakir

As in the documentation, The goal of fakir is to provide fake datasets that can be used to teach R.

Installation and Loading

fakir can be installed from Github (fakir isn’t available on CRAN yet)

# install.packages("devtools")
devtools::install_github("ThinkR-open/fakir")
library(fakir)

Use-case: Clickstream / Web Data

Clickstream / Web Data is one thing a lot of organizations use in analytics these days but it’s hard to get your hand on some clickstream data since no company would prefer sharing theirs. There’s a sample Data on Google Analytics Test Account but that may not serve you any purpose in learning Data science in R or R’s ecosystem.

This is a typical case where fakir can help you

library(tidyverse)
fakir::fake_visits() %>% head()
## # A tibble: 6 x 8
##   timestamp   year month   day  home about  blog contact
##   <date>     <dbl> <dbl> <int> <int> <int> <int>   <int>
## 1 2017-01-01  2017     1     1   352   176   521      NA
## 2 2017-01-02  2017     1     2   203   115   492      89
## 3 2017-01-03  2017     1     3   103    59   549      NA
## 4 2017-01-04  2017     1     4   484   113   633     331
## 5 2017-01-05  2017     1     5   438   138   423     227
## 6 2017-01-06  2017     1     6    NA    75   478     289

That’s how simple is to get a sample Clickstream (tidy) data with fakir. Another good thing to mention is, If you look at the fake_visits() documentation, You’ll find it that there’s an argument that takes seed value which means, you are in control of randomizing the data and reproducing them.

fake_visits(from = "2017-01-01", to = "2017-12-31", local = c("en_US", "fr_FR"), 
    seed = 2811) %>% head()
## # A tibble: 6 x 8
##   timestamp   year month   day  home about  blog contact
##   <date>     <dbl> <dbl> <int> <int> <int> <int>   <int>
## 1 2017-01-01  2017     1     1   352   176   521      NA
## 2 2017-01-02  2017     1     2   203   115   492      89
## 3 2017-01-03  2017     1     3   103    59   549      NA
## 4 2017-01-04  2017     1     4   484   113   633     331
## 5 2017-01-05  2017     1     5   438   138   423     227
## 6 2017-01-06  2017     1     6    NA    75   478     289

Use-case: French Data

Also, in the above usage of fake_visits() function you might have noticed another attribute local which can help you select French data instead of English. In my personal opinion, This is crucial if you are on a mission of improving Data Literacy or Democratising Data Science.

fake_ticket_client(vol = 10, local = "fr_FR") %>% head()
## # A tibble: 6 x 25
##   ref   num_client prenom nom   job     age region id_dpt departement
##   <chr> <chr>      <chr>  <chr> <chr> <dbl> <chr>  <chr>  <chr>      
## 1 DOSS… 31         Const… Boul… <NA>     62 Pays … 44     Loire-Atla…
## 2 DOSS… 79         Martin Norm… Cons…    52 Alsace 67     Bas-Rhin   
## 3 DOSS… 65         Phili… Géra… <NA>     28 Poito… 86     Vienne     
## 4 DOSS… 77         Simon… Cour… Plom…    29 Île-d… 91     <NA>       
## 5 DOSS… 59         Rémy   Dela… <NA>     18 Picar… 02     Aisne      
## 6 DOSS… 141        Astrid Dumo… Ingé…    35 Nord-… 62     Pas-de-Cal…
## # … with 16 more variables: gestionnaire_cb <chr>, nom_complet <chr>,
## #   entry_date <dttm>, points_fidelite <dbl>, priorite_encodee <dbl>,
## #   priorite <fct>, timestamp <date>, annee <dbl>, mois <dbl>, jour <int>,
## #   pris_en_charge <chr>, pris_en_charge_code <int>, type <chr>,
## #   type_encoded <int>, etat <fct>, source_appel <fct>

In the above example, We’ve used another function fake_ticket_client() of fakir that helps us in giving a typical ticket dataset (like the one you get from ServiceNow or Zendesk)

Use-case: Scatter Plot

So, the rant that I made at the start of this post about iris (Don’t mistake me: I’ve got huge respect for the scientists who created this dataset, it’s just that the wrong / over-usage of it which I don’t appreciate), Now we can overcome with fakir’s datasets.

fake_visits() %>% 
  ggplot() + geom_point(aes(blog,about, color = as.factor(month)))
## Warning: Removed 47 rows containing missing values (geom_point).

(Perhaps, Not a good scatter plot to show Correlation but hey, you can teach scatter plot without plotting Petal Length and Sepal Length)

Summary

If you are in the business of teaching or likes experimenting and don’t want to use cliched datasets, fakir is a very nice package to get to know. As the author of fakir’s package mentions in the description, charlatan is another such R-package that helps in generating meaningful fake data.

References

• fakir – Github
• fakir – Documentation