Since the beginning of this game, there were a lot of complaint about the replay's discrepancies with actual play. but I still guess that the unit's deployment points (coordinates, timing) are same... and AI plays the 'replay' again. Still, these days I lose too many times...quite strange..... but when i see the replay, strangely 4 out of 5 games wins!! ofcourse! because this successful deployment was my knowhow i've been playing for more than hundreds of attack. (currently my rankings are 690~750) how come there can be this big discrepancies? is this like some conspiracy sort of thing or just a bug? can't understand anyway..
IMHO re this and HTH!--> Been getting/hearing this and most of the feedback I got was, this is a known issue/bug within the game with regards to the replay. Many scenarios have been laid down maybe to help the root cause but still Devs are still on it for the solution. One thing I learned from my previous game, same game theme, same issue with regards to the replay of battles that "the replays are not actually a video or captured motion motion of the battle". That maybe the reason why its not perfectly captured all the every details of the battle on a replay like what we are expecting for example we capture a video of something and re/play it over and over again and the same thing happens. Think PP(or other game) is doing something like a log of the battle which probably a series of text/codes as logs and upon replay the system would do what's written/coded but somewhere in between slips happened(so issues/bug happen) so what you expect and knew happened doesn't show on the replay. But I know Devs are looking into this, maybe its just a matter of prioritization and time what known issues/bugs needs to address first.
This will supposedly change in the next update, but from what we've been told, all replays currently do is this: Records where and when each troop is deployed Records when each skill is triggered. Everything else is left up to the game's AI. While any discrepancy can affect the replay outcome, for me, the biggest changes come when an important troop (i.e. WD) that survived the battle is killed in the replay. Or, troops which dispersed in the battle, cluster together in the replay and get splattered by a mortar. Those two things can make a huge difference in the outcome.
I lost by 45% in 2 games but in replay it was 63% win... and 1 game lost by 45% but won by 61%, which had 16% discrepancies and i think this is huge problem proving that there are some logic problem. If the replay keeps the deploy timing, skill used timing, then it should have some consistency
I disagree. There are far too many variables in the AI for one to ever get a reliably consistent Replay/Battle result, especially when tracking only two data points. Think of it like a game of Chinese Whispers/Telephone. Each small error makes the replay and battle diverge a bit more. Eventually, they are nothing alike. You can see this in some replays by watching them multiple times. Even those don't always play out the same.
I'd think the AI should do the same as the original attack, unless there is some randomness intentionally built into AI (i.e. Plug 2+2 into a calculator and you'll always get 4. Computers don't make math errors unless overheated, etc). Perhaps the times recorded are rounded/slightly off which affects the replays. Nevertheless I'm sure replay issues will be corrected someday.
There is randomness built into the troop and defense AI. Not sure about randomness in the timing, but I don't see why that couldn't be included as well. The devs said that Replays should be more accurate as of the next update. Here's hoping.
yesterday i've won by 100% but replay was 50%... having this huge discrepancies should mean alot to everybody, because, if this is about AI stuff, this is too random result..... and this means it's nothing to do with your knowhow, experience of the game. this proves that it's about the 'luck of AI each time' not your 'skill' and this explains why I lose so many times though I think I now know how to deploy units to win.. this means I can't make any attack strategy videos if it is not consistent for everybody.
Does Midoki claim randomness in AI? To me I don't see any significant randomness. Basically bombers head straight towards objects and blow thru walls, all other units head towards closest target, except thief to resources and brutes towards defenses aiming at them. Yoon, I can typically pick out what I did wrong on my failed attacks; don't despair, it does appear to be a game of skill more than randomness.
Pick a troop or two, maybe a gunner or Brute and watch them during the battle. Then watch them during the replay. They do not always take the same path from their boat to the first objective --or anywhere else, for that matter.
Kelani, By following a unit, you've essentially said the replay and battle don't match. My question was on the battle AI whether Midoki states there is randomness built in? Not trying to argue any point but there are loads of conjecture and personal opinions on these forums (someone posted their favorite farming level on Mondays - seriously, is someone that observant to distinguish Monday from Tuesday?). I hadn't seen anything specifically on battle and replay AI being the same AI, but I easily could have missed it being a relative noob on forums. *** I haven't seen anyone post that watching replays yields different results watching the same battle replay multiple times. That would confirm randomness, but they seem to have the same errors, arguing against randomness. If true that replays repeat same paths then I assume different AI on battle vs. replay. **** Just watched the same replay three times straight, same paths for troops each time. Big logical argument AGAINST randomness, if random AI the replays would be different each time as well.
Sorry I was unclear there. To my knowledge, Midoki has never commented on the subject. I'm just going off my observations, and what I know of how game AI works. To create more realism and to stop players from learning and exploiting the "pattern", there is almost always some variability implemented.
random would be an absolute necessity for a game such as PP. if multiple replays are identical but different to the original battle, I'd surmise that; [1] random() results are being stored for replays (logical n necessary), and [2] some random() results are being neglected; ie, not captured; hence the difference between the original battle n the replays. just brain storming here... it's also possible that the battle data is being streamed back to the server during the battle, and some of the data is being dropped. if that were the case, theyd either need a better streaming method or a cache/upload-later method.
The discrepancy is out of hand at times. For example, I had a log showing that I was 2 star 96%ed in an attack. Yet my replay shows a 1 star 43% attack. In what weird world is this replay of any use?
Just to advertise my own post, I wrote a guide for the developers over here on how to attack the problem: http://www.midoki.com/forum/index.php?threads/how-to-debug-replay-determinism-issues.4994/ I just hope someone sees it that can take advantage of it. I agree it's very frustrating that the replays don't match up to what actually happened...it makes it somewhat pointless to try to improve your base design since you never know what the real weaknesses are.
The only issue I have with the replays being off is that they make me look like I was kicked in the head by a mule, thought I was calling 911 for help, but was actually deploying troops. All of my replays are much worse than my actual attack...but this may be evidence to prove one important point...AI is not as skilled/powerful as natural intelligence. If terminator had the AI from PP, the film would have ended with his first encounter with the bikers (when he gets his clothes). On the serious side of this conversation, the only times I see major changes is after splash damage is applied: mortar, mine, or ground pounder. It may have something to do with the random trajectory/results of those explosions.
