Frequently Asked Forum Questions | ||||
Search Older Posts on This Forum: Posts on Current Forum | Archived Posts | ||||
Re: DMR Study Update #3 (aka for Hoovaloov) | |
Posted By: Hoovaloov | Date: 1/24/12 5:10 a.m. |
In Response To: Re: DMR Study Update #3 (aka for Hoovaloov) (RC Master) : It was a two leveled critique. One that you should treat it fairly if you're : going to include it, and two that even if you do, I don't think it has : much relevence anyway. One does not preclude the other. Ok, I can respect that. Also, if I remember correctly, you were saying that it was unfair to call a 12% and 4% increase at short and medium range significant, while saying a 7% decrease at long range insignificant, right? I still say that the gains made at short and medium range will manifest themselves more often in the real world than the decrease at long range. Perhaps my logic is flawed, but try this analogy on for size (also realize that these numbers come from my friend Joel's equations, they don't seem to match the "real world" numbers I got at all, so I'm not talking about the actual percentages as much as why I think similar percentages can be called more or less significant): How noticeable the change is is directly related to the probability of you getting that result in the first place. Let's imagine a coin with pace written on one side and spam on the other. But on the spam side, I glued a lead weight to it (long range). That coin will hardly ever land spam up, no matter how many times I flip it. The lead weight (long range) means pace wins an overwhelmingly large majority of the time. Now what if I cut a small piece off of the lead weight, but leave most of it stuck to the spam side. I just increased the chances of the coin landing spam, right? (this is what the TU does, slightly increase the spam chances) But it will still hardly ever land spam up because there's still a lot of the lead weight (long range) glued to the spam side. You won't see the increase manifest itself because the percentage is so low anyway (22%). This example shows the difference an extra 7% spam kills makes at long range. It's not very noticeable because the coin lands pace up an overwhelming majority of the time even though spam is increased by 7%. Now imagine that I glue a bit of wood (short range) to the pace side of a new coin. The coin now wants to land spam up, but you can imagine it landing pace up sometimes now, but mostly spam. Now I'll cut off most of the wood stuck to the coin, improving the chances of pace by 12%, you'll really start to notice it because the coin is almost evenly weighted again (53%-47% spam-pace). This example shows the difference an extra 12% can make between 65% pace and 53% pace. It's pretty noticeable because when you increased your chances by 12%, you going from large majority of spam wins to almost a 50-50 toss up. The change will manifest itself more often. I realize that this is just an analogy, so it doesn't quite do it justice. But I think the logic makes sense. Winning at short range by pacing goes up 12%, but that 12% puts us very close to 50-50 odds with Joel's numbers. We just went from a small minority winning percentage to almost dead even. But at long range, the TU's 7% increase in spam wins just turns spam wins from being an extremely rare occurrence to a very rare occurrence. Once again, please realize that I am using Joel's numbers because that's the context you used to point out that I was being "biased." I think we both agree that Joel's numbers don't match up with the real world very well. ---------------------------------------------------------------------------- : Hold up. Firstly, the quoted margin of difference is rather a retreat from
Yes, I was saying my conclusion is stronger because my frame of reference was Joel's numbers from Update #2. Update #3 of the study provided us with percentages that, interestingly enough, closely matched the winning percentages from the original spreadsheet. Since these numbers (+10% short, +6% med, -1% long) ended up matching my conclusion more closely than Joel's "theoretically ideal" numbers (+12% short, +4% med, -7% long), I was pointing out that the conclusion had gotten stronger since Update #2. Sorry for the confusion. I was comparing Update #3's results with Update #2's and observed that Update #3's results were stronger. Also, I talked to Joel quite a while about his equations. I asked him why his equations came up with such a different result than my "1000 random samples" test. He said the difference is due to how much variance there is in the empirical data. He thought the two numbers would converge somewhere as more data was added to the sample set. In other words, his equations don't work very well with a small sample size. But that's all we have to work with, for now. ---------------------------------------------------------------------------- : Could you talk a little about how you made that? I looked at the spreadsheet
I suppose that wasn't very clear, since I don't think Google Docs uploads macros. I coded myself a custom macro in Excel to do the heavy lifting. 1) Random numbers are generated in those columns.
That's where I get the +10% at short, +6% at med, and -1% at long range results from. And these are much more "real world" results because this is as close as it gets to a real 1v1 with the numbers. So I think these empirical data sets paint us a better picture of what's really going on than my friend Joel's "theoretically ideal" results. What do you think? It's like we have two decks of 25 cards, one for pace and one for spam. Then we each take a deck, shuffle it, and play through all 25 cards. I tally up how many times in the game my cards beat yours. Then we reshuffle and play again for 1000 times total. Then I average my wins at the end. And do this for all six scenarios. ---------------------------------------------------------------------------- : Odds are at short range that spamming will still win, even if the rate is
If you're asking what the threshold of detectibility is, or how much of an increase is worth it, then honestly, we haven't defined it yet. But let's give it a shot. Let's say we play a game to 25 kills, a 1v1. We each will only fire our DMR one way: I'll pace and you'll spam all the time. According to my data, at short range, in 100% bloom, I have a 28% chance of winning any given encounter. This means in a game to 25, you should win 25-10. Now let's do the same thing in 85% bloom, at short range. Now I have a 38% chance of winning any given encounter. This means in a game to 25, you should win 25-15. I got 5 more kills this time. Is that noticeable/significant? Yeah, definitely. Big improvement for me. What about medium range? In 100% bloom, I have a 63% chance of wining a given encounter. So in a game to 25, I should win 25-15. Now in 85% bloom, I have a 69% chance of winning a given encounter. So in a game to 25, I should win 25-11. You got 4 less kills. Big improvement for me. What about long range? In 100% bloom, I have a 85% chance of winning any given encounter. So in a game to 25, I should win 25-4. Now in 85% bloom, I have a 84% chance of winning a given encounter. So in a game to 25, I should win 25-5. You just got one more kill. Not a big deal, really. So yes, at short and medium range, the increase in winning percentage is very noticeable, but not so much at long range since the 100% and 85% bloom result in virtually the same pacing winning percentage. ---------------------------------------------------------------------------- : But wasn't the premise of the article the rate difference between the two?
I don't really understand your point here. This is in context of the DMR being fired properly. My point is that I was defining pacing as a very strict thing - all shots must wait for fully reset reticule. This leads to much slower kill times than "optimal" DMR firing (i.e. a few quick bodyshots, short pause, bodyshot + quick pacehot or something like that). So optimal DMR firing would probably get close to 50-50 with spammers in 100% bloom, and maybe even overtake them in 85% bloom. The only way to figure this out would be to figure out the total time for an optimal DMR kill. Then we can replace that time with all the paced times in the spreadsheet, and run the simulation 1000 (or 25,000 :) times. So let's try to figure that out. Since spamming is the most likely to win at close range, let's start with that. I just went back and watched my videos - in both 100% bloom and 85% bloom short range tests, the first 4 spam shots always hit the target, popping the shields. The problem spamming had at short range was landing that last pacehot because the bloom was maxed out by then. A spam/max ROF fire shot is 12 frames in both 100% and 85% bloom. For the first 3 bodyshots, that's 36 frames. Then after the 4th bodyshot, we let the reticule fully reset, which takes 15 frames in 100% and 13 frames in 85%. Also, add the 2 frames it takes the last shot to register. So that means in 100%, an "optimal" kill would be 36+15+2=53 frames. For 85%, it's 51 frames. Before I plug that into my Battle Simulation program, what do you think of these numbers. Reasonable? ---------------------------------------------------------------------------- : Spammers have larger targets and can recover more quickly (max ROF - 12
85% bloom takes away the one shot "cushion" the spammer has. He can no longer miss once and recover to win the encounter against the pacer like he could in 100% bloom. Therefore, 85% bloom rewards the accuracy of pacing more than 100% does by punishing spamming. It's much riskier to spam in 85% bloom because the pacer's ROF is faster. He can overtake the spammer if he misses. ---------------------------------------------------------------------------- : On the same token let me quote you
I think I was looking at the whole picture, which is that the results of experiment is getting to the underlying changes the TU makes to the win probabilities of pacing and spamming. I removed almost all the variables, so pretty much all that's left is the TU changing things. Any changes such as aiming at a different target, factoring in player accuracy, movement, all these things are affected equally by the TU's changes. You were pointing out very specific 1v1 encounters to make a point, I was saying that the large trend of the TU is that pacing wins are increased at short & medium range, and at long range is virtually the same as 100% bloom. But maybe I totally don't understand your point. :) ---------------------------------------------------------------------------- : Part in bold: for the DMR? The DMR only becomes a 4 shot if you've taken any
Sorry - my hypothesis was two-fold. The first being that the NR is the gun to use in the TU. It clearly has some funky bullet magnetism stuff going on, and the reticule resets faster than the DMR anyway. It's a pacehot machine in the TU. The second part of the hypothesis was that the best way to fire either gun (DMR or NR) would be to spam 3 bodyshots (if medium/long range), pause, 1 bodyshot + 1 pacehot. Or spam 4 bodyshots (if short) needed to pop the shields, and then go for the pacehot. That's how I try to play anyway. ---------------------------------------------------------------------------- Sorry for the crazy long post. I'll upload the maps and gametypes for you later today. How should I let you know when they're in there? Gamertag = TTL Hoovaloov Also, I think I'm going to re-write the article in a new post sometime soon that excludes all the early calculations that we both agreed aren't accurate. The way it stands now is kind of mess since I updated it three times. :) Good discussion, RC. It's been fun fleshing out this study. As iron sharpens iron...
|
|
Replies: |
The HBO Forum Archive is maintained with WebBBS 4.33. |