Econometrics and Free Software by Bruno Rodrigues.

RSS feed for blog post updates.

Follow me on twitter, or check out my Github.

Watch my youtube channel.

Have a question about one of the blog posts? Chat here with me.

The year of the GNU+Linux desktop is upon us: using user ratings of Steam Play compatibility to play around with regex and the tidyverse

R

I’ve been using GNU+Linux distros for about 10 years now, and have settled for openSUSE as my main operating system around 3 years ago, perhaps even more. If you’re a gamer, you might have heard about SteamOS and how more and more games are available on GNU+Linux. I don’t really care about games, I play the occasional one (currently Tangledeep) when I find the time, but still follow the news about gaming on GNU+Linux. Last week, Valve announced something quite big; it is now possible to run Windows games on GNU+Linux directly from Steam, using a modified version of Wine they call Proton. The feature is still in Beta, and Valve announced that they guarantee around 30 games to work already flawlessly. Of course, people have tried running a lot of other games, and, as was to be expected from Free Software and Open Source fans, GNU+Linux gamers created a Google Sheet that lists which games were tried and how they run. You can take a look at the sheet here.

In this blog post, I will play around with this sheet. This blog post lists some {tidyverse} tricks I find useful and use often. Perhaps these tricks will be useful to you too! Let’s start by loading the needed packages:

library(tidyverse)
library(magrittr)
library(readxl)

Since I’m lazy and don’t want to type the whole name of the file I’ll be using some little regex:

steam <- read_excel(Sys.glob("Steam*"), sheet = "Main", skip = 2)

glimpse(steam)
## Observations: 8,570
## Variables: 9
## $ SteamDB   <chr> "LINK", "LINK", "LINK", "LINK", "LINK", "LINK", "LINK"…
## $ Game      <chr> "64", "1849", "1982", "1982", "am Weapon: Revival", ".…
## $ Submitted <chr> "5 days ago", "12 days ago", "11 days ago", "11 days a…
## $ Status    <chr> "Garbage", "Platinum", "Gold", "Platinum", "Platinum",…
## $ Notes     <chr> "Crashes with a debug log", "Plays OK.", "Gamepad supp…
## $ Distro    <chr> "Arch (4.18.5)", "Manjaro XFCE", "Gentoo AMD64 (Kernel…
## $ Driver    <chr> "Nvidia 396.54 / Intel xf86-video-intel (1:2.99.917+83…
## $ Specs     <chr> "Intel Core i7-7700HQ / Nvidia GTX 1050 (Mobile)", "Ry…
## $ Proton    <chr> "3.7 Beta", "3.7-4 Beta", "3.7-4 Beta", "Default", "3.…

Let’s count how many unique games are in the data:

steam %>%
    count(Game)
## # A tibble: 3,855 x 2
##    Game                                                                   n
##    <chr>                                                              <int>
##  1 .hack//G.U. Last Recode                                                2
##  2 $1 Ride                                                                1
##  3 0rbitalis                                                              1
##  4 10 Second Ninja                                                        4
##  5 100% Orange Juice                                                     17
##  6 1000 Amps                                                              3
##  7 12 Labours of Hercules VII: Fleecing the Fleece (Platinum Edition)     1
##  8 16bit trader                                                           1
##  9 1849                                                                   1
## 10 1953 - KGB Unleased                                                    1
## # … with 3,845 more rows

That’s quite a lot of games! However, not everyone of them is playable:

steam %>%
    count(Status)
## # A tibble: 8 x 2
##   Status       n
##   <chr>    <int>
## 1 Borked     205
## 2 bronze       1
## 3 Bronze     423
## 4 Garbage   2705
## 5 Gold       969
## 6 Platinum  2596
## 7 Primary      1
## 8 Silver    1670

Around 2500 have the status “Platinum”, but some games might have more than one status:

steam %>%
    filter(Game == "100% Orange Juice") %>%
    count(Status)
## # A tibble: 5 x 2
##   Status       n
##   <chr>    <int>
## 1 Bronze       5
## 2 Garbage      3
## 3 Gold         2
## 4 Platinum     6
## 5 Silver       1

More games run like Garbage than Platinum. But perhaps we can dig a little deeper and see if we find some patterns.

Let’s take a look at the GNU+Linux distros:

steam %>%
    count(Distro) 
## # A tibble: 2,085 x 2
##    Distro                                         n
##    <chr>                                      <int>
##  1 <NA>                                           1
##  2 ?                                              2
##  3 "\"Arch Linux\" (64 bit)"                      1
##  4 "\"Linux Mint 18.3 Sylvia 64bit"               1
##  5 "\"Manjaro Stable 64-bit (Kernel 4.14.66)"     1
##  6 "\"Solus\" (64 bit)"                           2
##  7 (K)ubuntu 18.04 64-bit (Kernel 4.15.0)         2
##  8 (L)Ubuntu 18.04.1 LTS                          1
##  9 18.04.1                                        1
## 10 18.04.1 LTS                                    2
## # … with 2,075 more rows

Ok the distro column is pretty messy. Let’s try to bring some order to it:

steam %<>%
    mutate(distribution = as_factor(case_when(
        grepl("buntu|lementary|antergos|steam|mint|18.|pop|neon", Distro, ignore.case = TRUE) ~ "Ubuntu",
        grepl("arch|manjaro", Distro, ignore.case = TRUE) ~ "Arch Linux",
        grepl("gentoo", Distro, ignore.case = TRUE) ~ "Gentoo",
        grepl("fedora", Distro, ignore.case = TRUE) ~ "Fedora",
        grepl("suse", Distro, ignore.case = TRUE) ~ "openSUSE",
        grepl("debian|sid|stretch|lmde", Distro, ignore.case = TRUE) ~ "Debian",
        grepl("solus", Distro, ignore.case = TRUE) ~ "Solus",
        grepl("slackware", Distro, ignore.case = TRUE) ~ "Slackware",
        grepl("void", Distro, ignore.case = TRUE) ~ "Void Linux",
        TRUE ~ "Other"
    )))

The %<>% operator is shorthand for a <- a %>% f(). It passes a to f() and assigns the result back to a. Anyways, let’s take a look at the distribution column:

steam %>%
    count(distribution)
## # A tibble: 10 x 2
##    distribution     n
##    <fct>        <int>
##  1 Ubuntu        6632
##  2 Arch Linux     805
##  3 Solus          175
##  4 Debian         359
##  5 Fedora         355
##  6 Gentoo          42
##  7 Void Linux      38
##  8 Other           76
##  9 openSUSE        66
## 10 Slackware       22

I will group distributions that have less than 100 occurrences into a single category (meaning I will keep the 5 more common values):

steam %<>%
    mutate(distribution = fct_lump(distribution, n = 5, other_level = "Other")) 

steam %>%
    count(distribution)
## # A tibble: 6 x 2
##   distribution     n
##   <fct>        <int>
## 1 Ubuntu        6632
## 2 Arch Linux     805
## 3 Solus          175
## 4 Debian         359
## 5 Fedora         355
## 6 Other          244

Let’s do the same for the CPUs:

steam %<>%
    mutate(CPU = as_factor(case_when(
        grepl("intel|i\\d|xeon|core2|\\d{4}k|q\\d{4}|pentium", Specs, ignore.case = TRUE) ~ "Intel",
        grepl("ryzen|threadripper|tr|amd|fx|r\\d|\\d{4}x|phenom", Specs, ignore.case = TRUE) ~ "AMD",
        TRUE ~ NA_character_
    )))

steam %>%
    count(CPU)
## Warning: Factor `CPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`
## # A tibble: 3 x 2
##   CPU       n
##   <fct> <int>
## 1 Intel  5768
## 2 AMD    2319
## 3 <NA>    483

And the same for the GPUs:

steam %<>%
    mutate(GPU = as_factor(case_when(
        grepl("nvidia|geforce|3\\d{2}|nouveau|gtx|gt\\s?\\d{1,}|9\\d0|1060|1070|1080", Specs, ignore.case = TRUE) ~ "Nvidia",
        grepl("amd|radeon|ati|rx|vega|r9", Specs, ignore.case = TRUE) ~ "AMD",
        grepl("intel|igpu|integrated|hd\\d{4}|hd\\sgraphics", Specs, ignore.case = TRUE) ~ "Intel",
        TRUE ~ NA_character_
    )))

steam %>%
    count(GPU)
## Warning: Factor `GPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`
## # A tibble: 4 x 2
##   GPU        n
##   <fct>  <int>
## 1 Nvidia  6086
## 2 AMD     1374
## 3 Intel    413
## 4 <NA>     697

I will also add a rank for the Status column:

steam %<>%
    mutate(rank_status = case_when(
        Status == "Platinum" ~ 5,
        Status == "Gold" ~ 4,
        Status == "Silver" ~ 3,
        Status == "Bronze" ~ 2,
        Status == "Garbage" ~ 1
    ))

Now, what are the top 5 most frequent combinations of Status, distribution, CPU and GPU?

steam %>%
    filter(!is.na(CPU), !is.na(GPU)) %>%
    count(Status, distribution, CPU, GPU) %>%
    mutate(total = sum(n)) %>%
    mutate(freq = n / total) %>%
    top_n(5)
## Selecting by freq
## # A tibble: 5 x 7
##   Status   distribution CPU   GPU        n total   freq
##   <chr>    <fct>        <fct> <fct>  <int> <int>  <dbl>
## 1 Garbage  Ubuntu       Intel Nvidia  1025  7443 0.138 
## 2 Gold     Ubuntu       Intel Nvidia   361  7443 0.0485
## 3 Platinum Ubuntu       Intel Nvidia  1046  7443 0.141 
## 4 Platinum Ubuntu       AMD   Nvidia   338  7443 0.0454
## 5 Silver   Ubuntu       Intel Nvidia   650  7443 0.0873

Unsurprisingly, Ubuntu, or distributions using Ubuntu as a base, are the most popular ones. Nvidia is the most popular GPU, Intel for CPUs and in most cases, this combo of hardware and distribution is associated with positive ratings (even though there are almost as many “Garbage” ratings than “Platinum” ratings).

Now let’s compute some dumb averages of Statuses by distribution, CPU and GPU. Since I’m going to run the same computation three times, I’ll write a function to do that.

compute_avg <- function(dataset, var){
    var <- enquo(var)
    dataset %>%
        select(rank_status, (!!var)) %>%
        group_by((!!var)) %>%
        mutate(wt = n()) %>%
        summarise(average_rating = weighted.mean(rank_status, (!!var), wt, na.rm = TRUE))
}

Let’s see now if we can rank distribution by Steam play rating:

compute_avg(steam, distribution)
## # A tibble: 6 x 2
##   distribution average_rating
##   <fct>                 <dbl>
## 1 Ubuntu                 3.03
## 2 Arch Linux             3.05
## 3 Solus                  3.03
## 4 Debian                 3.01
## 5 Fedora                 3.07
## 6 Other                  3.16

How about for hardware?

compute_avg(steam, GPU)
## Warning: Factor `GPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`

## Warning: Factor `GPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`
## # A tibble: 4 x 2
##   GPU    average_rating
##   <fct>           <dbl>
## 1 Nvidia           3.07
## 2 AMD              2.90
## 3 Intel            3.01
## 4 <NA>            NA
compute_avg(steam, CPU)
## Warning: Factor `CPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`

## Warning: Factor `CPU` contains implicit NA, consider using
## `forcats::fct_explicit_na`
## # A tibble: 3 x 2
##   CPU   average_rating
##   <fct>          <dbl>
## 1 Intel           3.03
## 2 AMD             3.06
## 3 <NA>           NA

To wrap this up, what are the games with the most ratings? Perhaps this can give us a hint about which games GNU+Linux users prefer:

steam %>%
    count(Game) %>%
    top_n(10)
## Selecting by n
## # A tibble: 10 x 2
##    Game                              n
##    <chr>                         <int>
##  1 Age of Empires II: HD Edition    43
##  2 Borderlands                      39
##  3 DiRT 3 Complete Edition          32
##  4 DOOM                             62
##  5 Fallout: New Vegas               45
##  6 Grim Dawn                        34
##  7 No Man's Sky                     40
##  8 Path of Exile                    35
##  9 Quake Champions                  32
## 10 The Elder Scrolls V: Skyrim      46

I actually laughed out loud when I saw that DOOM was the game with the most ratings! What else was I expecting, really.

If you found this blog post useful, you might want to follow me on twitter for blog post updates.