Study on artillery

[Shinma 296]

Ok, I'd suggest we pause the responses for a second and address a definition issue that keeps getting touched on but ignored. 

 

 

 

We don't have a unified definition of 'campiness' or camping

 

 

How does this sound?

 

 

Camping: Play which cedes initiative to the other side through reduced movement; (sic) sitting back and letting the other person move.

 

Campiness: The degree to which both teams engage in camping. High 'Campiness' = minimal aggressive movement

 

 

If we go with this definition, it gives me an idea of a much easier to measure baseline.

 

(Someone who doesn't suck at math as bad as I do, feel free to rewrite this less retardedly)

 

  (1 / Y) / X=(D / M)    

Where:

X= Campiness

D= Damage 

M= Distance travelled (in meters)

Y = Number of Artillery per battle

 

 

So the theory that arty reduces camping would be: Damage to distance ratio will be lower as the number of arty increases.

[SWG112 299]

You could in theory average non LT and non SPGs spots to get an idea, since a more aggressive team will generally have a more even distribution of spots among reular tanks. Thig is that depends a bit too much on tank and skill level.

Distance traveled imho it's wrpng thinking about it deeper. I travel much more in arty games purely because of the evasive maneuvres whenever i'm spotted in the open, while in a non-arty game (1 or less and not too big caliber) i might just sit and spam shots...

[RichardNixon 835]

Some results. I gave up on trying to normalize by map: There's simply too many maps given the per-game noise and my replay count. Tiers came out with some reasonably solid numbers though:

Tier 1, count = 506, lifetime = 203, mileage = 538
Tier 2, count = 4004, lifetime = 209, mileage = 689
Tier 3, count = 7771, lifetime = 201, mileage = 853
Tier 4, count = 13099, lifetime = 214, mileage = 884
Tier 5, count = 24483, lifetime = 236, mileage = 951
Tier 6, count = 34516, lifetime = 254, mileage = 1002
Tier 7, count = 35631, lifetime = 285, mileage = 1062
Tier 8, count = 48190, lifetime = 310, mileage = 1091
Tier 9, count = 25377, lifetime = 323, mileage = 1186
Tier 10, count = 18134, lifetime = 346, mileage = 1382

The counts are (non-arty) tanks rather than battles. Apparently tier 10 players drive around a lot.

 

So then I normalized by tier and ran the calculation with arty count as the variable:

0 arty, count = 63420, lifetime = 1.021, mileage = 0.977
1-2 arty, count = 83049, lifetime = 0.998, mileage = 0.999
3-4 arty, count = 33927, lifetime = 0.989, mileage = 1.020
5-6 arty, count = 29646, lifetime = 0.976, mileage = 1.027
7-8 arty, count = 1200, lifetime = 0.960, mileage = 1.036
9-10 arty, count = 469, lifetime = 0.885, mileage = 1.046

So lifetime apparently goes down with arty count, and mileage goes up. The differences are small compared to those between tiers. It's probably statistically significant, but I can't be bothered to figure that out.

 

Source is 8000 replays from wotreplays. I also ran the same test on a couple of single-player 1000 and 1500 replay batches and got very similar results.

[Shinma 296]

Can you try a sample tank or two for the damage/distance test from that data pull? 

 

For tier X, say 140 and e100? 

 

 

I just realized that replays as data source for this might have some bias. Doesn't replay parsing require full battles? So Instant cancer won't process. 

[RichardNixon 835]

Can you try a sample tank or two for the damage/distance test from that data pull? 

 

For tier X, say 140 and e100? 

 

E-100: Count = 1826, lifetime = 359, mileage = 1155

Obj140: Count = 765, lifetime = 327, mileage = 1835

 

 

The wotreplays source is massively biased because it's selective, although because 29 out of 30 samples are not the player it doesn't necessarily make that much difference to the results. It depends what you're testing. Relative tests (like 0 vs 2 arty games from the same source) tend to be useful, but absolute tests (eg average tier for the player, win margin) are not.

 

The single-player batches both come out at ~4/5 with results sections. Again, because only 1 out of 30 samples are the player, the bias is likely to be small, and it arguably makes it more representative of what the average WoTLabs poster will see.

[Shinma 296]

E-100: Count = 1826, lifetime = 359, mileage = 1155

Obj140: Count = 765, lifetime = 327, mileage = 1835

 

 

The wotreplays source is massively biased because it's selective, although because 29 out of 30 samples are not the player it doesn't necessarily make that much difference to the results. It depends what you're testing. Relative tests (like 0 vs 2 arty games from the same source) tend to be useful, but absolute tests (eg average tier for the player, win margin) are not.

 

The single-player batches both come out at ~4/5 with results sections. Again, because only 1 out of 30 samples are the player, the bias is likely to be small, and it arguably makes it more representative of what the average WoTLabs poster will see.

Sorry, I should have been more specific Basically it's two count. Count 1 = tank. count 2 = arty category 1 = Damage Category 2 = Milage 

 

results should be for both count 1 and 2 and just for count 1, 

[WildSudowoodo 49]

So the numbers show that distance traveled increases when the number of SPGs in game increases? If so, then back to what I said on the first page.. What about distance traveled in the beginning stages of a game? Obviously the more SPGs hiding in the back at the end of the game, the more people will drive around looking for them.

[RichardNixon 835]

Sorry, I should have been more specific Basically it's two count. Count 1 = tank. count 2 = arty category 1 = Damage Category 2 = Milage 

 

results should be for both count 1 and 2 and just for count 1, 

 

Well, now I definitely have no idea what you're talking about.

 

You can't split the single-tank data by arty count because there would be too few samples. It'd just be noise.

[Shinma 296]

Well, now I definitely have no idea what you're talking about.

 

You can't split the single-tank data by arty count because there would be too few samples. It'd just be noise.

Are you using a python parse for this? If so, it'd be catching all replays where "typecompdescr" = 9489 (OR 16897)  , throwing every replay that doesn't match.

 

For those that do parsing "DamageDealt" and "Milage"  only for matching typecompdescr.

 

Then sorting said results by Vehicletype to put those with arty in category 1, and those without into category 2. 

 

 

So the end result will be just the damagedealt / milage for 9489 in results 1 (arty) and damagedealt / milage for 9489 in result 2.

 

 

See how many samples come up just on sorting by typecompdescr 9489.

[attaker 3]

I think the best way to study the campiness of a game would be average distance traveled by non-arty and non-scout tanks during the first five minutes as well as damage dealt during the first five minutes.

[RichardNixon 835]

This is going to be pretty noisy:

Obj140 only:
0 SPG, count = 213, lifetime = 334.864, mileage = 1797.770
1-3 SPG, count = 328, lifetime = 331.189, mileage = 1838.430
4+ SPG, count = 224, lifetime = 314.741, mileage = 1865.138

E-100 only:
0 SPG, count = 442, lifetime = 386.629, mileage = 1204.882
1-3 SPG, count = 860, lifetime = 355.335, mileage = 1148.664
4+ SPG, count = 524, lifetime = 343.347, mileage = 1123.040

T10s by class uses rather more of the data:

T10 MT only:
0 SPG, count = 1301, lifetime = 337.917, mileage = 1843.135
1-3 SPG, count = 2305, lifetime = 327.243, mileage = 1882.719
4+ SPG, count = 1502, lifetime = 320.634, mileage = 1915.111

T10 HT only:
0 SPG, count = 1859, lifetime = 357.838, mileage = 1265.350
1-3 SPG, count = 3621, lifetime = 342.048, mileage = 1250.899
4+ SPG, count = 2387, lifetime = 324.729, mileage = 1219.953

T10 TD only:
0 SPG, count = 1279, lifetime = 377.515, mileage = 1089.573
1-3 SPG, count = 2299, lifetime = 373.457, mileage = 1094.013
4+ SPG, count = 1581, lifetime = 367.528, mileage = 1100.277

So the heavies take the biggest lifetime hit, while the mediums account for most of the mileage gain. Heavy mileage also drops as arty count increases, so they're apparently taking most of the punishment. Arty apparently doesn't make much difference to tier 10 TDs.

[Shinma 296]

I think the best way to study the campiness of a game would be average distance traveled by non-arty and non-scout tanks during the first five minutes as well as damage dealt during the first five minutes.

If you read back through the thread, that method is extremely hard to process large samples of. It'd take a full rewrite of the existing parsing scripts and replay packet analyzers for damage done in the first 5 minutes. Distance travelled would pose a huge fucking problem since you'd be calculating raw co-ordinate changes per tank and ignoring map scales. 

 

short version, it's not doable.

[MaxL_1023 1,295]

That data makes sense - heavies camp, mediums chase and TDs barely move to begin with so they are static. 

[Shinma 296]

This is going to be pretty noisy:

Obj140 only:
0 SPG, count = 213, lifetime = 334.864, mileage = 1797.770
1-3 SPG, count = 328, lifetime = 331.189, mileage = 1838.430
4+ SPG, count = 224, lifetime = 314.741, mileage = 1865.138

E-100 only:
0 SPG, count = 442, lifetime = 386.629, mileage = 1204.882
1-3 SPG, count = 860, lifetime = 355.335, mileage = 1148.664
4+ SPG, count = 524, lifetime = 343.347, mileage = 1123.040

T10s by class uses rather more of the data:

T10 MT only:
0 SPG, count = 1301, lifetime = 337.917, mileage = 1843.135
1-3 SPG, count = 2305, lifetime = 327.243, mileage = 1882.719
4+ SPG, count = 1502, lifetime = 320.634, mileage = 1915.111

T10 HT only:
0 SPG, count = 1859, lifetime = 357.838, mileage = 1265.350
1-3 SPG, count = 3621, lifetime = 342.048, mileage = 1250.899
4+ SPG, count = 2387, lifetime = 324.729, mileage = 1219.953

T10 TD only:
0 SPG, count = 1279, lifetime = 377.515, mileage = 1089.573
1-3 SPG, count = 2299, lifetime = 373.457, mileage = 1094.013
4+ SPG, count = 1581, lifetime = 367.528, mileage = 1100.277

So the heavies take the biggest lifetime hit, while the mediums account for most of the mileage gain. Heavy mileage also drops as arty count increases, so they're apparently taking most of the punishment. Arty apparently doesn't make much difference to tier 10 TDs.

Again, still missing damagedealt. I'm not sure how lifetime/milage is testing anything other than lategame vs early game movement 

 

Also, cutting off at 3 arty is odd, is it arty/2 or 3 instances? In the latter, it makes no sense to look at 3 arty games, since it's the odd 1 higher tier vs 2 lower tier. 

 

 

 

All that said, how would we interpret the numbers?

 

Edit: 

 

Thought about it, this kinda shits on the "arty reduces camping argument" since there's an inverse correlation between mobility/milage and survival rate when arty is introduced.

 

Thanks Richard, in case I haven't said it yet, for the work on this so far btw.

[RichardNixon 835]

Again, still missing damagedealt.

 

Damage dealt is useless unless you adjust for available hit points, which is a small project in itself and highly error-prone. There's a clear correlation between available hit points and arty count.

 

I assume you don't want damage dealt as a camping metric, as I can't think of a situation where it would be superior to average lifetime.

 

 

Also, cutting off at 3 arty is odd, is it arty/2 or 3 instances? In the latter, it makes no sense to look at 3 arty games, since it's the odd 1 higher tier vs 2 lower tier.

 

No point in throwing away the data. 1 arty games are extremely rare anyway. I considered 3 arty games to be more similar to 2 arty games than 4 arty games, but it's not going to make much difference either way.

 

 

Thought about it, this kinda shits on the "arty reduces camping argument" since there's an inverse correlation between mobility/milage and survival rate when arty is introduced.

 

Lifetime isn't survival rate. If someone survives a short game, they'll have a lower lifetime than if they eventually died in a long game. If the average lifetime is lower, then the games are shorter and more brutal for the non-arty players.

 

You could reasonably argue that the mileage difference on the mediums is simply due to them having to move further to kill the arty. A result that may be overlooked is that tier 10 heavies actually move further per second (3.76m vs 3.5m) in the 4+ arty games.

 

My favourite theory is that a higher arty count simply leaves more space on the battlefield and that accounts for a large chunk of the effects. What's more likely to dissuade you from moving up: Possibly taking an RNG shot from arty, or the Waffle covering the space in front of you?

[Shinma 296]

Damage dealt is useless unless you adjust for available hit points, which is a small project in itself and highly error-prone. There's a clear correlation between available hit points and arty count.

 

I assume you don't want damage dealt as a camping metric, as I can't think of a situation where it would be superior to average lifetime.

Damage dealt/milage was the purpose of damage dealt. Hit points within a match are static, so damage/milage means as the ratio increases, the 'campiness' decreases.  

No point in throwing away the data. 1 arty games are extremely rare anyway. I considered 3 arty games to be more similar to 2 arty games than 4 arty games, but it's not going to make much difference either way.

It's making a breakpoint that obscures data. 1 Arty games are glitches, so there's not much to be statistically gained.

 

 

Lifetime isn't survival rate. If someone survives a short game, they'll have a lower lifetime than if they eventually died in a long game. If the average lifetime is lower, then the games are shorter and more brutal for the non-arty players.

 

You could reasonably argue that the mileage difference on the mediums is simply due to them having to move further to kill the arty. A result that may be overlooked is that tier 10 heavies actually move further per second (3.76m vs 3.5m) in the 4+ arty games.

If the map is smaller (ensk, himmels, kharkov, mines) the effect of arty is negligible. You're also talking about a phenomenon that heavies and TDs are going to be more likely to not engage in, since -logically speaking- faster tanks arrive sooner at further points.

My favourite theory is that a higher arty count simply leaves more space on the battlefield and that accounts for a large chunk of the effects. What's more likely to dissuade you from moving up: Possibly taking an RNG shot from arty, or the Waffle covering the space in front of you?

Once again, at the risk of overstating the obvious, arty will hit targets not in cover. Few cancer can hit over, or splash around obstacles. In your scenario, the waffle is both a known quantity, and simple burst left vs. HP calculations can be made; this before the most fundamental difference that direct fire is a two way street.

[RichardNixon 835]

Damage dealt/milage was the purpose of damage dealt. Hit points within a match are static, so damage/milage means as the ratio increases, the 'campiness' decreases.

 

Four things that damage dealt in a game tells you:

 

1. Hit points available in the game. Needs to be factored out. There is no easy way to compile a list of elite tank hp, which is why I need a bloody good reason to compile one.

 

2. Draw chance. Camping-related, but you can access the result directly without need for proxy variables.

 

3. Cap chance. Inversely related to camping. Maps that have high cap rates (eg Stalingrad, Highway) have relatively low damage dealt, for example.

 

4. Steamroll chance. Inversely related to camping. Maps with frequent one-sided games (eg Kharkov, Murovanka) have relatively low damage dealt.

 

Unless I'm missing something, it's a better camping indicator in the other direction, especially if you throw away the draws. Essentially, lower damage dealt in won/lost games means more cap wins and more steamrolls. Most players here probably don't think that cap wins and steamrolls are good things, but they're not camping either.

 

 

#1 If the map is smaller (ensk, himmels, kharkov, mines) the effect of arty is negligible.

 

#2 You're also talking about a phenomenon that heavies and TDs are going to be more likely to not engage in, since -logically speaking- faster tanks arrive sooner at further points.

#3 Once again, at the risk of overstating the obvious, arty will hit targets not in cover.

 

#1 Even if that was true, it doesn't matter. If you have to drive further on some maps to kill the arty, then the mileage numbers will come out higher.

 

#2 Which fits the evidence rather well.

 

#3 In which sense it's no different from any other class. Only the definition of cover changes.

 

The rest is mostly psychology. I don't personally recognise the description of arty as some deeply scary camping-inducer, more like one fewer tank to punch through before the free kills. Maybe that changes when you're playing tier 10 against XVM abusers. Still, it's the average player that counts for the numbers, and most of them barely look at the roster. If a unicum hides behind a rock while the pubbies are dying like flies, that's not going to come out as a camping game. If the argument is that "arty makes unicums camp", then you'd need a different test.

[Shinma 296]

Four things that damage dealt in a game tells you:

 

1. Hit points available in the game. Needs to be factored out. There is no easy way to compile a list of elite tank hp, which is why I need a bloody good reason to compile one.

WN8 isn't making the same distinction, I don't see a research design issue if we follow suite.

2. Draw chance. Camping-related, but you can access the result directly without need for proxy variables.

 

3. Cap chance. Inversely related to camping. Maps that have high cap rates (eg Stalingrad, Highway) have relatively low damage dealt, for example.

Both of those points would need further testing.

4. Steamroll chance. Inversely related to camping. Maps with frequent one-sided games (eg Kharkov, Murovanka) have relatively low damage dealt.

Low Damage in steamrolls is due to only one team putting meaningful damage.  

Unless I'm missing something, it's a better camping indicator in the other direction, especially if you throw away the draws. Essentially, lower damage dealt in won/lost games means more cap wins and more steamrolls. Most players here probably don't think that cap wins and steamrolls are good things, but they're not camping either.

I'll try to rephrase my hypothesis to hopefully clear up the confusion.

it's not damagedealt, but damagedealt in relation to distance travelled. The point of the thread was to try and get some empirical measurement of what effect arty has on camping. You've already pointed out the problems with distance travelled and survival time (small maps + steamrolls/doesn't say whether a person survived, or match duration).

By having another ratio to examine it, or a three-point result, it'll help to remove intervening variables.

The reason I'm trying to get data on damagedealt/Milage is that besides having ll the wnX data on damage, explanations for why damage to milage changes with respect to presence of arty seems to have far fewer alternative explanations than milage/lifetime. It's pretty rare that every person on either team will hard camp. The question is whether people perceive camping (less movement) as more rewarding (damage) as arty number increase.

 

 

#1 Even if that was true, it doesn't matter. If you have to drive further on some maps to kill the arty, then the mileage numbers will come out higher.

 

#2 Which fits the evidence rather well.

 

#3 In which sense it's no different from any other class. Only the definition of cover changes.

 

The rest is mostly psychology. I don't personally recognise the description of arty as some deeply scary camping-inducer, more like one fewer tank to punch through before the free kills. Maybe that changes when you're playing tier 10 against XVM abusers. Still, it's the average player that counts for the numbers, and most of them barely look at the roster. If a unicum hides behind a rock while the pubbies are dying like flies, that's not going to come out as a camping game. If the argument is that "arty makes unicums camp", then you'd need a different test.

re: 1, rarely is it the case that everyone on one side is surviving long enough to chase arty. 2/4 meds driving 400m for arty is still less distance than 4/4 driving 250m to overmatch. We've also got no other way to filter 'quality of milage' so either we throw out milage as a whole, or accept the flaws.

Re: 2 Does it? We know meds get higher milage. Chasing down arty only fits for 0- <1 arty; the number of places arty can and does hide is finite, and tends to be clustered. I get the feeling Redshire is mental touchstone on this.

Re: 3 except that we're not looking at arty's performance in this. The best way to shoot tanks in cover is to move where you can shoot them. (I'm sorry for the number of 'protip-level statements). If tanks are moving to shoot things in cover (IE, not camping) then both milage and damage should go up. If Tanks who follow this diaf, then damage will go down.

[GehakteMolen 2,063]

I quickly scanned over this topic, but this has a serious flaw, almost all camp maps are gone, (while arty is still annoying as hell), due to ``all the coridors`` camping is much less likely, compare el halluf, before rework a massive campfest (if there is arty) i bet like 50% longer average battle time, since rework however, its much less campy, and the campyness there is, is mostly around trying to move around and poking corners, its not about hiding behind a rock.

 

So while in theory interesting, this is perhaps a bit too late.

 

Regarding the calculation (not sure if its possible)

The most simple one is average battle time on same map with 0, 1, 2, 3, 4+ arty (on some maps it wont make difference (ensk) on other others, old El Halluf and murovanka beiing the clearest examples of mega-campfests, also campinovka and komarin (before reworks) where massive camp-partys if enough arty.

 

A better metric would be the dmg dealt / minute, i guess in camping games, most dmg is done either at the start, when campers kill rushers / unlucky red E100s, the mid part is emtpy (everyone hides) while late game there is a steep rise in dmg again (when all yolo forward and try to get some kills)

[RichardNixon 835]

WN8 isn't making the same distinction, I don't see a research design issue if we follow suite.

 

WN8 actually handles this problem implicitly, along with most other MM effects. If arty were not explicitly matched against other arty then the expected values would come out differently even if the method was unchanged. It's more of a problem trying to remove that effect from tank balance data.

 

Back to the subject, the obvious bias here is that a game with arty in it has a much lower hit point total. If you don't correct for that then you'll obviously get a result of more arty => lower damage dealt.

 

 

Low Damage in steamrolls is due to only one team putting meaningful damage. 

 

But a steamroll isn't a camping game, right? So you want damage dealt of the winning team, not the average? That's much simpler but clearly even less useful. There are now three cases:

 

1. Draw -> low damage. Can be measured directly anyway.

2. Cap win -> low damage, but not camping.

3. Other win -> all damage. Purely dependent on the losing team's starting hp.

 

 

You've already pointed out the problems with distance travelled and survival time (small maps + steamrolls/doesn't say whether a person survived, or match duration).

 

I don't know why you think that's a problem with lifetime. A rapid 15-0 thrashing isn't a camping game. Neither is a similarly rapid game where almost everyone dies. If anything it's a demonstration of why lifetime is a far better parameter than anything else suggested. Note that map variation is only noise, not systematic bias, assuming that arty are just as likely to appear on any map.

 

The only parameter I can think of that might be an improvement is some sort of median survival time, so for example you calculate when the 10th player dies. This removes any potential bias from arty-hunting.

 

 

Edit: Something else that might be worth measuring is how the effectiveness of players in various skill bands changes with arty count, so for example whether unicums perform better or worse in games with more arty. Would probably need per-tank hp for that though, as the sample's too small to use winrate.

[Shinma 296]

Uh, I forgot to hit send on a more lengthy response.

 

Alright, I see your point; I agree median survival time would be a decent filter (though per map would be better), or survival time/battle duration to filter any outliers. 

[attaker 3]

Would it be possible to find the average distance traveled per minute and average damage done per minute (all for non arty and non scout tanks)

 

Also I liked the idea of damage dealt per distance traveled mentioned earlier

[Shinma 296]

Would it be possible to find the average distance traveled per minute and average damage done per minute (all for non arty and non scout tanks)

 

Also I liked the idea of damage dealt per distance traveled mentioned earlier

Being the one who pitched the idea, I guess I'll help Richardnixon by fielding it as an act of contrition. 

 

So, the issues that damage/distance brings in is that there are too many competing explanations it fails to eliminate from its assumptions. 

 

Case 1: Map is Mines, damage/distance will be higher. 

Case 2: Pretty much every single encounter, damage/distance will be lower due to capout.

 

If you filtered by map, by class it could be a reasonable measurement, but the results would be so small that they'd be statistically irrelevant. (like 20-30 per band.) 

 

 

That said, it might be worth looking into putting something like damage/distance in wn9, where sample sizes are far larger than the 9500 replays parsed in this case. 

[Servios 467]

I quickly scanned over this topic, but this has a serious flaw, almost all camp maps are gone, (while arty is still annoying as hell), due to ``all the coridors`` camping is much less likely, compare el halluf, before rework a massive campfest (if there is arty) i bet like 50% longer average battle time, since rework however, its much less campy, and the campyness there is, is mostly around trying to move around and poking corners, its not about hiding behind a rock.

This is basically what's happened. It's difficult to measure it simply because the good spots to do it are progressively being removed leaving players with less options.

[A_Lethal1 1]

Throwing in another variable.  When a match has artillery in it does scouting aggressively discourage camping? Does passive scouting encourage camping in those matches with artillery.   I am thinking that there may be a correlation between both the number of scouts and how they are played with artillery in a match.