Renegade Stats for 2021

[Itai795]

This is going to be a long post...

I had some spare time this week and decided it will be a good opportunity considering we just started a new year to take a look at the saved game stats we have here for 2021 and see if we can have some interesting conclusions from it.

Also, English is not my native language, sorry for any mistakes that I probably made and don't have the patience to fix.

Edit: didn't realize it while I was writing the post, but I divided it into different pages. Not sure if I like it. But make sure you press the next/previous page buttons to see everything.

Topics for each page:

• Data
• Player Count Timeline
• Map Counting
• Match Duration
• Win Rates
• Win Types

 

---

Data

The file contains a data point for each match we had in the server, starting 1/1/21 (there are older files somewhere at that ftp server). In each data point we have:

• Date and time of the end of the match (I'm not sure what time zone it is, I think Eastern Time (New York). I'm also not sure if it changes between daylight saving and standard, it doesn't really matter though)
• Match length
• Mode (AoW / Marathon)
• Map
• Winning team (GDI / Nod)
• Win type (base destruction / ped beacon / score / server shutdown)
• Number of players at the end - We don't have data for how many players were during the match, so the data here is skewed (think that some players tend to ragequit towards the end of the match when a clear winner is already known). Keep it in mind when reading this topic.
• Score for each team

 

---

Players Count Timeline

First question we'll answer is how many players we have along the day, week and year. Starting with counting number of players (again, at the end of the match) when binning them into 24, 1-hour bins. Then, taking the 25, 50 and 75 percentiles (50 percentile is also called the median).

How exactly are we interpreting this graph and percentiles in general? Take for example the blue dot at the 15th hour, with the value of 5. It means that 25% of the matches that ended between 15:00-15:59 had less than 5 players. Similarly, 50% of the games that ended between 19:00-19:59 had less than 9 players.

For the following three timeline graphs, I'm going to omit all games that ended outside 14:00-21:59, where the median is less than 6 players.

Similar graph, but the binning is done by day of week:

An example how to read it: 75% of the games on Friday (between 14:00-21:59) had less than 22 players. Saturdays are out busiest. Friday is, surprisingly at least for me, busier than Sunday. Wednesday is the weakest day.

Another similar graph, with binning a weekly binning (dates here present the first day (Monday) of each week):

We had a drop at the beginning of July and we might seeing the start of returning to what we had at the beginning of the year.

---

Before continuing, I'm going to focus only on games that ended with more than 6 players. I don't want small games, where teams might be highly unbalanced (depending on actual players, afkers etc.), would affect the statistics we'll see further on. I also removed all maps that were played 15 or less times.

In addition. I also removed obscure game modes (capture the flag, domination, we had 10 and 3, respectively) and matches that ended due to server shutdown (177).

---

Map Counting

---

Match Duration

Following graph is an histogram graph, where we see the relative frequency of game duration. Each bar represents here 2 minutes.

Read it as ~6% (0.06) of the games ended between 6 to 8 minutes. The bump at the 44-46 minute bin is of course, the AoW time limit. The data for that graph is derived from all the maps. I looked at similar graphs specific for each individual map, most of them look the same: some maps tend to have their "end time" peak earlier or later than 6-8 as we see above, but the shape is quite similar. However, Field and Under are noticeable different, and tend to last much (much..) longer:

Now, instead of looking at histograms, we are going to look at the median match duration of all the maps, according to the number of players, for the two game modes (AoW, marathon). Before doing that, here are the number of matches played for each map, mode, and number of players (where the latter are grouped to bins:6-10 players, 11-15, 16-20, 21-30, 31-40 etc.)

Why is it important to show the count values? They would help us to understand how "significant" the match duration values in following graphs are for each map/mode/player combination. I'm going to be much more confident in the median duration of marathon Complex with 6-10 players than in the value of AoW Gobi 31-40 players because the first combination was played 122 times, while the latter combination was played only 2 times. I'm also going to drop all combinations that were played less than 8 times because it's too prone to be skewed by an outlier match.

So, again, following two graphs are the median duration of matches of different map / mode / players combo:

Same map again, this time without Field and Under, to make the colorbar useful for the other maps.

More players there are, games are longer. Under and Field act differently than the other maps. For the other maps, marathon games usually last less than 45 minutes, but I also thing that they last a bit longer than the AoW counterparts.

---

Maps that Fill (and Empty) the Server (Or: STOP LEAVING WTF)

Following graph shows what's the change in number of players between the end of a previous match to the current match, according to the currently played map. The bars represent the standard deviation. Green stands for marathon, red for AoW.

For example, when Field RxD marathon is on, it ends with (most likely) between 15 additional players to 3 less players than its predecessor map. Field RxD AoW fills the server a bit less than that- +11 to -4. Noticeable results: Field and Under fills the server (duh); Hourglass mara as well (surprise); Deth River mara too. Islands and Islands RxD (which are favorable maps) empties the server (maybe because their following maps in rotation are Field, IIRC); Mesa and Snow AoW are also not doing too well. Overall (to the right), mara maps fills the server a bit more than AoW.

---

Win Rates

Win rates for AoW and marathon for each map, ordered by % rate. The mean rates for each mode are written at the graph titles.

It's been known for a long time that Nod has better winning rates. I was surprised to see that it is actually more true with marathon. GDI are better with mara Walls flying and AoW Walls, but not with AoW Walls flying and mara Walls.

Let's reorder these two graph and see whether changing a mode from mara to AoW significantly changes the win rate for each time. I'm only considering the significant changes where the change was more than 5%. I ordered it from the most significant change to the left, and least one to the right:

What about win rate according to players count?

More players there are, more likely GDI is to win games. It's mostly true in AoW games, less in mara.

---

Win Types

Lastly, win type (base destruction, score or ped beacon) by map. Separating AoW and marathon. Starting with count numbers (for that map / win type combo), follows with % of all the played matches (under that combo).

Looking at the marathon graphs, almost all the maps end almost always with building destruction. Only Glaciers flying and Sand have some chances to end with ped beacon. I guess the mara matches that ended with high score are due to a vote. For AoW, building destruction is still the most likely win type, but is much less likely for Field and Field RxD (50%) and Under and Under RxD (66%). Glaciers flying has a relatively high chance of ped beacon win (25%).

We can also look whether number of players change the win type probabilities (again, starting with counting the number of matches):

Clearly we can see a trend that the more players in the game, more likely it to end with high score, while both base destruction and ped beacon decreases. Notice the 41-50 and 51-60 columns have really low number of played matches (12 and 1), not enough to be reliable.

Can't really say anything with confidence about marathon games because base destruction is the dominant winning type regardless to players count. However, we might see that high score is more often with high amount of players, maybe because games are so stalemated that only a vote saves the day, and we can also see that ped beacon is less likely with more players.

Edited January 16, 2022 by Itai795

[shaitan]

Thanks for reminding me...changed that so 2021's is here now.

[Vultima]

Woah this is really cool, what did you use to collate all the data?
and the graphics produced, what software is that?

[masscarriers]

You say that you're surprised that Nod wins more in mara than AOW but actually it makes a lot of sense especially on stalemate maps like Field and Under. Whereas GDI can hold their own for quite long and win by points in AOW, it's a lot more difficult for them to actually go and kill the base than it is for Nod in mara. How many time have we seen a strong base siege with meds/mammys lasting for quite long, only for them to lose the game because of a sneaky stank rush or because artys pushed through and blew the GDI base in mere minutes.

It's really interesting to notice how GDI wins Walls more often than Nod as opposed to basically every other maps. Is it because it's harder for them to advance with tech'd artys? They're exposed from so many sides on that map once they get out of their own base, including from up the mountain...

[Baseballs]

17 hours ago, masscarriers said:

It's really interesting to notice how GDI wins Walls more often than Nod as opposed to basically every other maps. Is it because it's harder for them to advance with tech'd artys? They're exposed from so many sides on that map once they get out of their own base, including from up the mountain...

In my experience the NOD team during Walls is easily 20-30% sbh, which means that GDI often dominates the battlefield and scores more points. However, as the graph shows, NOD actually wins more Wall games in mara mode; meaning that GDI has trouble destroying NOD's base, and in the long run NOD's chances of winning will increase because of sneaky SBH ped nukes or well executed stank rushes.

 

Edited January 9, 2022 by Baseballs

[Baseballs]

Thanks for designing these statistics Ita. Very interesting!

RX-maps seem to have a slight advantage for NOD, except for Canyon & Island in AoW. Maybe this is because of sbh+pimp weapons or NOD's lack of a gunner character?

I would also be interested to see if player count impacts 'win type'. Less players, more ped wins?

Same for player dynamics: does participation of @MateNone or @Lausetincreases the winning chances of the opposite team, as is often jokingly suggested? 🤣 Or maybe find out which player combinations have the most / least amount of wins? I don't know of that kind of detailed information is available, but it could generate some interesting data and funny banter.

 

 

[Itai795]

22 hours ago, Vultima said:

Woah this is really cool, what did you use to collate all the data?
and the graphics produced, what software is that?

I simply downloaded the data from here, I have no idea who wrote the bot the writes that file (I thought you are :) )

Generally speaking, I used python to analyze and plot the data. More specifically (if somebody wants to know), pandas is the analytical package (which is great for tabular data like what we have here), and holoviews with bokeh backend to produce the graphs.

 

20 hours ago, masscarriers said:

You say that you're surprised that Nod wins more in mara than AOW but actually it makes a lot of sense especially on stalemate maps like Field and Under. Whereas GDI can hold their own for quite long and win by points in AOW, it's a lot more difficult for them to actually go and kill the base than it is for Nod in mara. How many time have we seen a strong base siege with meds/mammys lasting for quite long, only for them to lose the game because of a sneaky stank rush or because artys pushed through and blew the GDI base in mere minutes.

It's really interesting to notice how GDI wins Walls more often than Nod as opposed to basically every other maps. Is it because it's harder for them to advance with tech'd artys? They're exposed from so many sides on that map once they get out of their own base, including from up the mountain...

I guess you are right. I had the impression that GDI would have more wins with mara over AoW because although they usually have hard time at the beginning, where Nod's arties+techs is a cheap and effective method, but only if GDI survives that and gets enough money for meds, havocs and perhaps orcas they are able to counter, and eventually win with meds+hotties and mammies. I guess that's true for maps such as Volcano, Mesa, Canyon, Complex etc. and less for Field and Under. You can actually see at the duration-players-map colorful tables that for these maps, game mode doesn't really affect the game duration, meaning whatever team that wins, wins regardless to time limit. For actual long mara maps, you might be right with the difficulty GDI is having to hold a siege and Nod can strike a surprise sneaky attack.

 

1 hour ago, Baseballs said:

Thanks for designing these statistics Ita. Very interesting!

RX-maps seem to have a slight advantage for NOD, except for Canyon & Island in AoW. Maybe this is because of sbh+pimp weapons or NOD's lack of a gunner character?

I would also be interested to see if player count impacts 'win type'. Less players, more ped wins?

Same for player dynamics: does participation of @MateNone or @Lausetincreases the winning chances of the opposite team, as is often jokingly suggested? 🤣 Or maybe find out which player combinations have the most / least amount of wins? I don't know of that kind of detailed information is available, but it could generate some interesting data and funny banter.

 

 

player count vs win type: interesting idea. I'm showing here the AoW graphs, and added both AoW and mara to the post. Marathon graphs are not very interesting, in AoW however there is a significant trend of more players = more likely for high score win.

 

player dynamics: yea I was thinking about it too, but unfortunately I don't think we log that data. Perhaps we save the IRC logs and I'll be able extract the data from there, @shaitan, do we have that?

Edited January 9, 2022 by Itai795

[Fardin]

Holy crap this was pretty impressive! Nicely done, you should publish a paper! :D  I also like how you included "stop leaving WTF!?!?!?"  xD
As baseballs has already mentioned, player dynamics would be pretty interesting. (especially extreme cases like April, having 200+ kills, Saurontwo or GPM with their super sneaking skills or rambo6 having only 20 points after 3h gaming lol) but I guess that would be very difficult data to obtain, let alone to work with.

Maybe the veteran points/ranks is something that could be worked with (Is that data even collected?) Though it's not a perfect representation of a players contribution, I feel like it certainly correlates somewhat. Also the teams combined veteran points: do they correlate with the winning rates? How many times does the team with more higher ranking players/points still lose the match? Stuff like that.

What about amount of vehicles/tanks? Since Renegade = Tank Game, I wonder how the amount of tanks correlate with victories (I assume there will be a very high correlation). Same could be said about the amount of useless snerips or sbhs. Using the Gamelog, I could imagine that it could be possible to collect data on how many tanks were brought, no?  

I wish there was a way to see, how the statistics are for the strategies that lead to victory, (classic mass tank rush vs arts and techs, sbh nuke or c4, tactical fake nukes and rushes, sneaky stanks/techs/hotties etc. ) but I know how there is pretty much no way to obtain that data except for actually recording all game footage and chat log to decide, which would be insane. (Maybe someday in the future someone will have enough spare time and no life to go through all of aprils or Irans streams on youtube and do that xD).

Lastly a personalized winning/losing rate would be pretty cool, like a sheet with all active player accounts and their respective winning losing percentages for each map and fraction as well as if they won after they joined or not lol but I understand how that has the same problems as the point with player dynamics, since that data isn't collected. Of course it doesn't matter for us two, since we won anyways, our winning rate is 100% 

Still, what you created here is pretty amazing, I never expected this much detail! I thought it will only be like a single bar plot with winning rates of the two fractions for each map. Do you do a lot of statistics/programming in your spare time/job?

[Itai795]

59 minutes ago, Fardin said:

Holy crap this was pretty impressive! Nicely done, you should publish a paper! :D  I also like how you included "stop leaving WTF!?!?!?"  xD
As baseballs has already mentioned, player dynamics would be pretty interesting. (especially extreme cases like April, having 200+ kills, Saurontwo or GPM with their super sneaking skills or rambo6 having only 20 points after 3h gaming lol) but I guess that would be very difficult data to obtain, let alone to work with.

1. Maybe the veteran points/ranks is something that could be worked with (Is that data even collected?) Though it's not a perfect representation of a players contribution, I feel like it certainly correlates somewhat. Also the teams combined veteran points: do they correlate with the winning rates? How many times does the team with more higher ranking players/points still lose the match? Stuff like that.

2. What about amount of vehicles/tanks? Since Renegade = Tank Game, I wonder how the amount of tanks correlate with victories (I assume there will be a very high correlation). Same could be said about the amount of useless snerips or sbhs. Using the Gamelog, I could imagine that it could be possible to collect data on how many tanks were brought, no?  

3. I wish there was a way to see, how the statistics are for the strategies that lead to victory, (classic mass tank rush vs arts and techs, sbh nuke or c4, tactical fake nukes and rushes, sneaky stanks/techs/hotties etc. ) but I know how there is pretty much no way to obtain that data except for actually recording all game footage and chat log to decide, which would be insane. (Maybe someday in the future someone will have enough spare time and no life to go through all of aprils or Irans streams on youtube and do that xD).

4. Lastly a personalized winning/losing rate would be pretty cool, like a sheet with all active player accounts and their respective winning losing percentages for each map and fraction as well as if they won after they joined or not lol but I understand how that has the same problems as the point with player dynamics, since that data isn't collected. Of course it doesn't matter for us two, since we won anyways, our winning rate is 100% 

Still, what you created here is pretty amazing, I never expected this much detail! I thought it will only be like a single bar plot with winning rates of the two fractions for each map. Do you do a lot of statistics/programming in your spare time/job?

Thanks mate :)

1. I really doubt vet poinst are logged anywhere besides at the instantaneous game. I'm ready to be surprised, but waiting for Shai or someone else who knows what we have and we don't.

2. I think that tank data can be obtainable from the IRC logs if they are saved, since every tank purchase and destruction shows up at the admin channels. I'm not sure about infantry purchases.

3. Theoretically, it's possible to guess (but you can't be certain) from the number and type of vehicles whether a rush is going on. Maybe combine with with chat logs, but that would be a real project. Not saying it has to be done manually, it's possible just to search for key words.. just theoretically speaking, I doubt it can lead to reliable results..

4. Yep so if we logged (or want to log that in the future), I can see how a different type of ladder can be created, considering what players are present at the game.

 

Yes doing these kind of stuff is part my job lol. It's much more fun here when it's about a game I love, and it also helps when you have a good intuition and preliminary assumptions about the insights would look like (like more players lead to longer games, points are irrelevant for mara games and so on)

[Jor3llBR]

Wow great stuff man, very impressive and interesting post. 🏆

[He4dSh0t]

So now can we actually discuss how to balance the game with very easy small subtle simple changes?

Because I have only numbers saved from only games I played and then ran some extrapolated numbers on how changes I would make would change things

[C123]

Awesome work @Itai795! Cool to see actual numbers behind what we (think) we already know.

Could you possibly normalize the "players gained" by game duration (so you get players joined per minute)? For example, a 2-hr Field/Under game might gain a bunch of players, but how much of that is because it's a 2-hour game (often in prime time) vs. people actually preferring Field/Under? Also, I think the player change graph would be more clear if you split it into two (one for aow, one mara).

Balancing ren would be a interesting project, but I wonder if Ren would be the same game without the Nod/stank advantage

 

[Forithow]

Yeah I really like the statistics that you collected here for 2021, good work Itai!

I do find it interesting that fan favourite maps such as Islands and Islands RxD will empty the server most often. I think that's due to the fact that Islands is after Field, so after a long match of Field, people would leave on the next map, which just happens to be Islands, so poor Islands is getting the short end of the straw :<

With the other maps that aren't doing as well, I wonder if we could replace them with a different map to see if they perform better, or mix the rotation up a bit to see if rotation order is what impacts why certain maps perform badly on player count? (Though this would probably be trivial since we can see that people leave on Islands after Field)

[N0]

Islands fan favorite? Pretty mid map tbh, only really 1 way for tanks to exit and enter which is bad map design, It's not close to the worst map in the game though.

[Itai795]

On 1/14/2022 at 12:26 PM, C123 said:

Awesome work @Itai795! Cool to see actual numbers behind what we (think) we already know.

Could you possibly normalize the "players gained" by game duration (so you get players joined per minute)? For example, a 2-hr Field/Under game might gain a bunch of players, but how much of that is because it's a 2-hour game (often in prime time) vs. people actually preferring Field/Under? Also, I think the player change graph would be more clear if you split it into two (one for aow, one mara).

Balancing ren would be a interesting project, but I wonder if Ren would be the same game without the Nod/stank advantage

 

I agree it would be more clear to separate the two errorbars (AoW and mara).. unfortunately in this particular case it's not easy to code to make it look good (needs to use a different approach than the one I used..)

So here is the normalized graph (numbers are player gain per minute):

And just for comparison, the not-normalized graph, which I already posted:

I'm not really sure what is the meaning of the normalized numbers to be honest, except the mean(+-standard deviation) of players gain per minute:

I don't think that you can conclude here that since Islands for example has a negative 0.25 players/minute score, we lost 2.5 players if it's on for 10 minutes. I think that the only right conclusion is that you lose (on average) 1.7 players when Island ends, and from the other tables I posted, Islands usually lasts about 8 minutes (didn't accurately check this number, just by looking at the tables). Take your typical Islands game (8 minutes) and add 30 minutes to it, I just don't think you are going to lose an addition 7.5 players (0.25*30) because of that.

Edited January 16, 2022 by Itai795