Minetest Servers plotted in 2D

[Hume2]

I took 90 Minetest servers and looked at their mods. There are 2179 different mods among these server. Each server is assigned a vector which contains 2179 booleans, one boolean for each of these mods. 1 means that the server has the mod, 0 means that the server doesn't have the mod. After that, I approximated all these vectors by a two-dimensional plane in a way that the sum of square distances to the plane are minimal. And then I projected the vectors onto the plane. The result is this:


It suggests that there are four clusters. I did the same for each cluster extra:

Top cluster:


Left cluster:


Bottom cluster:


Right cluster:


If you remove Must Test from the left cluster, it results in this:


I'm wondering if there is a meaningful interpretation of these plots. Can these clusters reflect player psychologies?

[Krock]

Hello there. I can do the same:


Generated using gnuplot and a custom application which does the cluster calculations.
Using 3D data: splot 'log.txt' u 2:3:4:1 w labels
Using 2D data: plot 'log.txt' u 2:3:1 w labels
Ignored mods: minetest_game mods, and extension mods in modpacks (such as mesecons_walllever)

For some reason gnuplot always crashes within 30s whenever I try to explore the graph. Great.

Graph only listing servers which contain many different mods:


UPDATE: Within a radius of 100 (of the global center) there's "minetesthosting.org" - as expected.

[Hume2]

Unfortunately, these 3D plots are unreadable because the distance from the screen isn't displayed.

A few questions:
How did you get the data? I needed to copy-paste the mod lists manually. Is there a way to query that automatically?
Some servers sometimes don't display the mod list for no apparent reason, including Tunnelers' Abyss. The mod list is sometimes displayed, sometimes isn't. It usually changes when the server restarts. Why does this happen and is there any way to fix it?

[sorcerykid]

Graph only listing servers which contain many different mods:

Interesting results. It's probably worth noting that going based on mod names alone might not be the most reliable indicator of uniqueness.

For example 34 mods on my server are original, so these can be distinguished easily:

• chat_history
• formspecs
• imagen
• livestats
• avatars
• signs_rx
• welcome
• trophies
• holoblock
• fnords
• launchpad
• snowfall
• xcommands
• toolbox
• stealth
• gopher
• athletics
• dataminer
• auth_rx
• digital
• navigator
• emoji_fx
• profiles
• cipher
• markup
• extra_doors
• macro
• physics
• registry
• ownership
• giftbox
• stopwatch
• inspector_rx,
  jt_mods

However, 16 mods were rewritten entirely from the ground up even though the legacy names were retained. So for all intents and purposes, they are original at this point:

• fireworks
• tnt
• bones
• inventory
• stairs
• mailbox
• snowdrift
• city_block
• protector
• wardrobe
• skins
• mt_seasons
• mobs
• teleport request
• teleports
• itemframes

And likewise, 4 mods happen to share identical names with mods released by other developers, even though the codebase is entirely unique: pathfinder, ambience, fireworks, inspector.

So I'm wondering, how this would effect the plots if these differences were factored in.

[Hume2]

When the mods have the same name, I threat them as the same. Even if their code was way different, the idea is basically the same.

Here you can look at the weights of the mods for both X and Y axes: https://gist.github.com/Hume2/ab3bada73 ... 0408f0df68
The unique mods usually have a small weight.

[sorcerykid]

The unique mods usually have a small weight.

I guess I'm confused what the plots are showing if not distinguishing between uniqueness vs commonality of mods. Is it number of mods? If so, I think that can be ascertained just by looking at the server list.

[Krock]

Unfortunately, these 3D plots are unreadable because the distance from the screen isn't displayed.

Indeed. I wanted to plot it in perspective view rather than isometric, but if there's an option within gnuplot, it's well hidden. Also 2D works, but there will be much less space to group similar servers.

How did you get the data? I needed to copy-paste the mod lists manually.

You can grab the JSON from the server list here: http://servers.minetest.net/list (just like Minetest does)

Some servers sometimes don't display the mod list for no apparent reason,

Server list or Minetest bug. It's constantly reported but the origin of this problem is not known yet.

Interesting results. It's probably worth noting that going based on mod names alone might not be the most reliable indicator of uniqueness.

Thank you. I am well aware that mod names alone do not indicate the effective uniqueness of the server. As of now, all mods are weighted equally to compensate the following extreme differences:
* Administrative tools do not change the player's impression of the server
* A custom server-provided texture pack is not a mod but adds a lot to the server's uniqueness

So I'm wondering, how this would effect the plots if these differences were factored in.

From your example, the server would be way further away from the center. "technic", "moretrees" and "advtrains" would also have a much higher impact due to the amount of content added to the server.

EDIT: About the grouping in my plots: The difference of two server's mods cause a higher distance, and a higher distance causes less impact of the mod difference. It's like a spring-steel physics simulation.

[twoelk]

most interesting stuff
Still don't understand the clusters
not anywhere near what I would expect

[Hume2]

You can grab the JSON from the server list here: http://servers.minetest.net/list (just like Minetest does)

Thanks, this will be helpful.

most interesting stuff
Still don't understand the clusters
not anywhere near what I would expect

I personally think that these four clusters match with classes in Bartle's taxonomy. Top cluster matches with socialisers, left one matches with killers, bottom one matches with achievers and right one matches with explorers.

[paramat]

And Must Test is just 'far out'.