The Oracle of Death

[maradine]

I'm working on the big MBT update now. Lightnings will be next.

[Goldoche]

Originally Posted by maradine

Those effects wash out over large numbers of players, and over large periods of time. The guy that can't stand crappy texture won't be coming back to pollute the numbers anymore.

Wrong. Those effects do not wash out over a large number players. All your data does is telling the preferred loadouts of the playerbase. If a player thinks a weapon is bad (it's not pretty enough, he heard bad things about it on the forums, it has just released (these are all are things that are affecting the playerbase as a whole, not as individuals)), he will rarely use it, lowering its KPU on the stat page.

Originally Posted by maradine

If you really, really want a per hour number, I re-iterate: Divide the kill number by what you see at the bottom of the readout, "hours"

Wrong. This is the time you used to sample a group, not the time players used their weapons.

The stats themselves are not bad in anyway, it's just people misinterpreting them that is annoying me.

[maradine]

Originally Posted by Goldoche

Wrong. Those effects do not wash out over a large number players. All your data does is telling the preferred loadouts of the playerbase. If a player thinks a weapon is bad (it's not pretty enough, he heard bad things about it on the forums, it has just released (these are all are things that are affecting the playerbase as a whole, not as individuals)), he will rarely use it, lowering its KPU on the stat page.

Only if he uses it in every sample period. He must make a kill with it to even register. This is exceedingly unlikely. You've said it yourself - he doesn't like it. He has no reason to use it. He will, on balance, not be included. I own a Flare. I didn't use it last week. I am not contributing to the denominator of Flare KPU. So, on balance, I am very comfortable disagreeing with you here.

Originally Posted by Goldoche

Wrong. This is the time you used to sample a group, not the time players used their weapons.

The stats themselves are not bad in anyway, it's just people misinterpreting them that is annoying me.

Equipped != Used. Playtime has its own considerable problems as a denominator. I have chosen not to use it. If my numbers were strongly disagreeing with traditional kills/playtime, I'd be more concerned with the objection. They are not, and I am not.

Look, you feel strongly about this. That's fine. Rather than argue about it, how about you find me a result I've run that doesn't make empirical sense, you show me how an alternate methodology would have scored it, and we talk about the difference and what it means. Otherwise, you're theorycrafting just the same as I am, and I only have time to code against one voice in my head right now.

Here - I'll give you a nice big one to find something to pick apart:

[maradine]

Been meaning to get back around to this one - enjoy!



Original doc: http://bit.ly/121hHBh

[typhaon]

SOE really needs to provide a functioning ES-AI secondary to NC.

I wish the Walker would be allowed a bit more of a downard tilt - making it a more viable option vs. infantry/vehicles... buff the Kobalt a bit... and do something to keep the Basilisk in between those 2.

The VS+NC 4Q stats for the E540 are interesting, especially compared to TR. I'd like to see how those look over time.

Vulcan seems very pedestrian on a Prowler, as opposed to on a Harasser. I know the weapons aren't exactly the same - but I assume this speaks to the platforms, available counters, and engagement ranges.

[maradine]

Originally Posted by typhaon

Vulcan seems very pedestrian on a Prowler, as opposed to on a Harasser. I know the weapons aren't exactly the same - but I assume this speaks to the platforms, available counters, and engagement ranges.

That was the biggest eye-opener for me. I was expecting complete apparent dominance as per the prior run. Granted, apples and oranges - I'll finish out the request backlog then re-do the Harasser set in the same interval.

Originally Posted by maradine

About 13 a second at the moment. Fluctuates with the global peaks from about 7 to about 25.

Must have been asleep at the wheel when I posted this. Current global kill velocity is coming out of the nightly trough at 750/(45+3)=15.6 kps. Actual nadir was 13.6. Peak is closer to 50 kps.

[Wahooo]

Originally Posted by Goldoche

The stats themselves are not bad in anyway, it's just people misinterpreting them that is annoying me.

I thought that is what statistics are for.

[Varsam]

Originally Posted by Goldoche

The stats themselves are not bad in anyway, it's just people misinterpreting them that is annoying me.

I have to agree with this, actually. While the numbers are novel and interesting, I'd never use them to support anything other than a claim about popularity. Attempting anything else (mostly weapon effectiveness/balance) would be dangerous - there seems to be no rational correlation between the two.

[Stomps]

Originally Posted by Varsam

I have to agree with this, actually. While the numbers are novel and interesting, I'd never use them to support anything other than a claim about popularity. Attempting anything else (mostly weapon effectiveness/balance) would be dangerous - there seems to be no rational correlation between the two.

Easy way to solve it, the Oracle could publish Kills Per Hour Equipped as well and let the users see both sides of the coin.

[Oracen]

Originally Posted by Varsam

I have to agree with this, actually. While the numbers are novel and interesting, I'd never use them to support anything other than a claim about popularity. Attempting anything else (mostly weapon effectiveness/balance) would be dangerous - there seems to be no rational correlation between the two.

I would read the statistics differently. I registered to reply to Exokhay, who had another interpretation of the statistics. Bastard wiped his post

I read the high KPU as a sign of weapon versatility, or possibly degree of mastery required (although the latter is far harder to prove, and is well in the realms of conjecture).

High KPU weapons tend to be weapons known for decent performance across multiple ranges. In the case of the TR, the highest KPU weapon was the T16 Rhino. This weapon is known for its ease of handling and good performance across all ranges (if not fantastic performance in any given scenario). It is, however, a niche choice. Contrast this to the MSW and TMG, weapons you'd expect your average TR player would tell you to get. They got low KPUs, from memory a bit over half of the T16. I'd argue this is due to players only using them in areas where they feel they get an advantage (real or otherwise, think close range with the MSW) and switching out when the situation changes.

This would result in 2 entries of unique killers with the one person, and would not be indicative of any balance issues regardless of the massive difference in KPU. The Vanu weapons at the top of the LMG list function, in many ways, like the T16. The greater control means that in close range, the losses of fire rate/damage can be mitigated by consistency of headshots. Without being an expert on VS/NC weapons (or even TR, for that matter ) it would seem the latest carbine pull would support my claim. With the exception of UBGL carbines, it would seem most of the versatile/ranged carbines (ranged being relative here) score higher than the close-ranged bullethoses which you would expect higher scores with.

To put it in numbers, you pull data from 2 people (let's condense the entire sample size to these two, just for shits and gigs). One uses the MSW and TMG, and gets 20 kills, in a ratio of 1:1. The other uses the T16 and gets 15 kills. Even though the former player got more total kills, he gives a KPU of 10 to the MSW and a KPU of 10 to the TMG. The second, surprise surprise, contributes a KPU of 15 to the T16. This doesn't indicate that the T16 is any more effective, nor that it's overpowered. Rather, this is a reflection that it can be utilised in more situations. I know most of the people in this thread will get this already, but this is more so everyone knows where my head is at before I'm misquoted. Blame 8 months of the official forums lol.

MAX weapons would seem to support this theory; Fracs are much more popular than the next most versatile weapon (Mercies)...and as someone who regularly deploys Fracs against squishies, I can say that I do it because I know I'll be able to hit targets regardless of engagement range. It's also the reason I use the T5 AMC and SABR as my go-to weapons. The lack of viable alternatives on the TR MAX is far more of a factor than any degree of OPness, although if people think Fracs are OP I'm not going to discuss it here. (More than happy to though, but I just don't want to derail a thread with my first post.) Comets, similarly, are accurate at all ranges and synergise well with ZOE. Mattocks? Weeelllll...they're more versatile than MOST NC MAX weapons, and despite the small number of users, score very highly. Skill is no doubt a factor here though, as they're still lacking in the versatility department.

If I may beg the Oracle, I think we'd find AR numbers would throw some additional light onto the problem, and I think we'd see a similar phenomenon; versatile weapons outperforming their specialised counterparts.

Exokhay suggested the high KPU of Terran vehicles with low LMG figures would suggest TR spend more time in their tanks. I would argue this, and this is where I bring in my "degree of mastery" conjecture comes out to play. I would suggest that Prowler shots, by nature of the dual-shot cannon, require a small, extra degree of handling skill. This would make adjusting to the handling profile of another set of cannons more difficult, and as such I believe players are more likely to stick with the one primary.

In my mind, this is kind of similar to my versatility argument, but approaching it from the other direction. The easier it is to switch between cannon ballistic profiles, the lower the KPU will be. One player, 2 unique killer IDs, one set of kills. I'm not saying TR weapons are uber1337/hXc, but they would naturally have an added hurdle to switching between the two. This may be enough to skew the numbers, but you'd need to start comparing weapon kills per ID as a percentage of total session time or some such to prove it. If anyone has this, I'd expect to see Prowlers spending 10-20% more time with one cannon than VS/NC who I think would switch it up more often.

The alternative is the Prowler's performance is just a factor of that higher base DPS, but I'd like to see recent K/D data pulled; I think suicide running may be vital there. Or it may be Striker dominance? Who even knows. Conjecture is fun

EDIT: WHOA. Wall of text. I apologise guys, apparently wine makes me ramble a bit.

[SonofSkz]

Two interesting points in the data are high level vanu are scoring better with the standard RL compared to the other factions by kpu which suggests at least that high level vanu will tend towards using the default more than the other two factions. I think this may be purely due to the fact that the lancer is useless against infantry, although that doesn't explain why the TR are behind there, maybe TR are using the Striker purely for its Area of Denial benefits allowing them to focus on shooting infantry with their lmgs? Another I thought was fun was the marauder was spanked in top tier kpu by the ppa suggesting in the hands of a good player the ppa can be very effective from the top of the tank, this is good because I'd like to see more vanu giving the ppa the props it now deserves

[Blahnu]

^The PPA is pretty amazing. What's interesting is the how on KPU, MBT-PPA beats MBT-Marauder but Harasser-PPA is outclassed by Harasser-Marauder. Not entirely surprising I guess, but worth noting. I'm curious about how the two weapons compare on the Harasser for experienced players, so I'm eagerly awaiting the 2nd pass on Harasser weapons.

The default launcher KPU for high BR Vanu is weird, I can't come up with many plausible explanations. For indoor/anti-MAX work a high BR player, Vanu or not, would likely own and pull a Decimator - ABR and KPU quartiles for the Decimator corroborate this. Maybe some high BR Vanu were trying to Aurax the S1 during the weekend.

Originally Posted by Oracen

If I may beg the Oracle, I think we'd find AR numbers would throw some additional light onto the problem, and I think we'd see a similar phenomenon; versatile weapons outperforming their specialised counterparts.

The Assault Rifle numbers were posted earlier, they're the first thing on page 9. I agree with you on the point that KPU is more an indication of versatility than anything, on infantry weapons at least.

The versatile ARs for the NC and TR - Gauss S (UBGL), T1S Cycler (UBGL), Carnage BR and TAR are the highest KPU ARs for those two factions. The Vanu UBGL is however less popular, with more preferring the H-V45. You'll notice that the Vanu use their short-range AR much more than the other two factions use theirs (GR-22/TRV).

I suspect a lot of Vanu medics use the NS-11A for a mid-range option - general consensus seems to be that the CME and Corvus are better suited for long-range engagements, with the H-V45 being forced in the 'versatile' role for those who don't own the NS-11. It is worth noting that the CME does have the highest KPU for the Vanu but there were few users.

[maradine]

Originally Posted by Blahnu

^The PPA is pretty amazing. What's interesting is the how on KPU, MBT-PPA beats MBT-Marauder but Harasser-PPA is outclassed by Harasser-Marauder. Not entirely surprising I guess, but worth noting. I'm curious about how the two weapons compare on the Harasser for experienced players, so I'm eagerly awaiting the 2nd pass on Harasser weapons.

Be careful about comparing numbers across early runs where I didn't explicitly pull timestamps and publish a duration, or published a different duration. KPU scales up over time - more slowly than kills, of course, but enough to eliminate direct comparison with the early stuff.

Originally Posted by Varsam

I have to agree with this, actually. While the numbers are novel and interesting, I'd never use them to support anything other than a claim about popularity. Attempting anything else (mostly weapon effectiveness/balance) would be dangerous - there seems to be no rational correlation between the two.

In this interpretation, how would you interpret the Striker? (post #142)

I'd also like to point out that, if we see KPU go up with BR quartile, for what is reasonably interpreted as skill/exp reasons, then we have independent confirmation that kpu reflects the effectiveness of the user/weapon pair. When we take the whole group, we are left with the weapon.

Put simply: if skill has an effect on KPU (it seems to), it reasonably follows that weapon effectiveness does as well, independent of the popularity of the weapon.

[Chaff]

.
Love this thread. I appreciate your time & the other factors at work that compelled you to start down this path, and continue to do so.

I also am interested in the methodolgy discussion(s) & the theoretical & assumptive reasoning required to try to objectively discern what the data as you choose to grab & share it means. Good stuff. Thanks again for your time & effort.

When you (oh great and Mighty Oracle of Death) get to Lightnings, I'd like to see the data non-compiled. I'm curious to see how close the numbers are between the 3 different empires. I suspect the 4th Empire players will help flatten the curve between the 3 substantially. May be minutiae to everyone else, but my twacked melon will find any difference(s) interesting.

Tanks again for all the fine data and the discussions that spring forth as a result.
.

[Aarth]

One thing you must consider with the MBT weapon stats, especially for the Prowler, is how the main tank cannon competes with the secondary for kills. The Prowler HEAT and HE are remarkably efficient against infantry, it may be so much so that Marauder gets pushed down a bit in its role. Another factor is that TR tanks may get more field time just due to having less threat from the sky, i.e. the massive effect the Striker seems to have on air play and game play in general.