1. Welcome to the Orange Islands! Please remember to read the rules. Come chat with us on Discord and enjoy your stay!
  2. The moderators of this forum are DHR-107, Codraroll, and NoCheese.
  3. Welcome to Smogon Forums! Please take a minute to read the rules.
  4. Click here to ensure that you never miss a new SmogonU video upload!

Battle Tree Discussion and Records

Discussion in 'Orange Islands' started by NoCheese, Nov 18, 2016.

  1. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    Also your pokemon's team or move's presence will greatly alter the sequence of those trainers popping up as well.
    But I do recognize that some people here that has a lot of streaks sometimes will lose to a battle to a team combo that I will meet before 10 or 20 battles. Either they are super lucky, they play on an old AI that prob didn't adapt as quickly, or they switch teams... :P
  2. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    Their problem is that they are not running Hone Claws Blaziken.
    Smuckem and Altissimo like this.
  3. GG Unit

    GG Unit
    is a Community Contributor

    Joined:
    Jun 30, 2008
    Messages:
    518
    This is categorically false and the equivalent of saying the AI cheats.


    This is also false. There is no need to disparage other players' teams based on your own inability to play the game.
    Codraroll, CTNC, JustinTR and 3 others like this.
  4. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    To be honest, wanting to be extremely pedant, depending on what data the game fetches when generating the AI trainers, there is indeed a slim possibility that the numeric IDs of the dex numbers, moves, natures, EVs, and plenty of other data could be affecting the trainers generated.
    Being more realistic, considering the stupidly high amount of combinations of movesets, dex numbers, natures, and what not, it's borderline impossible that even in the event that such data affects the trainer generation, it is actually "coded to counter".


    It is true though that we do lose to Azumarill-3 more often than we should, while usually novice players don't have problems with it.
    Altissimo likes this.
  5. paperquagsire

    paperquagsire

    Joined:
    Jan 24, 2017
    Messages:
    39
    Azumarill3 always beats my hone claws blazekin, which is unbeatable unless I see it. I saw it so therefore the AI cheats
    Altissimo likes this.
  6. Coeur7

    Coeur7

    Joined:
    Nov 28, 2016
    Messages:
    59
    Of course not specifically to me. You were referring to all players who switch teams after the early battles when you argued that they should be removed from the leaderboard, or at least should have the battles deducted from their streaks (with overtones of this being "a grey area" close to cheating), and more-or-less subtly suggested that they are less apt/creative at team-building or Tree-playing. I merely happen to have stumbled across how a syllogism is supposed to work, one time. (And a few more times, I've employed quantifiers.)

    To boot, Pencilcheck has gone even further than that in claiming (without even one concrete calculation or argument regarding specific teams and specific set1/2 threats -- which I find incredibly arrogant to people who actually spend hours on planning, because they enjoy it) that switching teams amounts to essentially cheating one's way past the first 20 battles, because the second team supposedly could never have gotten past that initial ceiling. Incidentally, I believe it was him who brought up that Durant cannot possibly win against Garchomp-1 and that the computer specifically draws sets that can counter one's strategy (an admirably optimistic stance about machine learning -- or human diligence!).

    There are, in fact, games whose deviously subtle decision-making algorithms choose randomly most of the time, but spot/target simple weaknesses on occasion (such as FTL on Hard difficulty, which I've heard has been discovered just recently -- by disassembling the source, the only way to actually prove this stuff), but that requires some effort to design. The lack of Battle Frontiers since gen5 despite e.g. the Factory being demonstrably a matter of 1-2 hours to code (given the framework of what's already in the game) makes me doubtful that Pokémon Sun/Moon are among these; design goals seem to differ between what I want from Pokémon and what the developers want. I've griped about that elsewhere. It's not quite sound reasoning, of course.

    Anyway, though, I proceed under the assumption that the RNG is not biased because the outcome is ultimately quite the same: both in team-building and playing, I have to insure myself against low-probability events that are bound to happen sooner or later. If the actual chance of QC activating was indeed 75% for the computer from battle 120 onward or whatever, this chance would still be nearly-indistinguishable from 20% over a sufficient number of battles battles if a certain such activation would spell doom to my team.

    In fact, the people who have spoken out against team-switching have apparently not realized that -- due to them also being interested in peak and not average performance of any team (otherwise, you would have scrapped your teams that got defeated several times before reaching 40; your premature conclusion about the Quick Claw also suggests this) -- even a team that struggles terribly against the likes of Bronzong-1, but razes the entire roster from battle 21 onward (obviously an exaggeration), only needs to overcome those supposedly critical early battles once to still achieve its streak all the same. Thus, in fact, a "blitz team" still achieves nothing but saving time under those assumptions, because surely it can't be that unlikely to draw 60 manageable set1/2s with horrible IVs, and proceed to sweep everything else.

    I'll speak from my own perspective and experience because those are the only ones I have. When I think something is wrong, I'll say it -- and expect to get the favour returned when I'm wrong about something.
    It's been said, but you're misunderstanding probability -- and perhaps don't realize that your video and suggested test precisely fit under "anecdotal evidence".

    QC activating 3 times in a row occurs once per 125 battles on average. It's actually more unlikely that an event of this probability doesn't occur once during your leaderboard streak. You'd need to record the outcome of all RNG-based events across a meaningful number of battles (1 is not meaningful) and compare this data against the expected probabilities. Which is not quite as easy to test, but excellent results tend to require some work.
    Last edited: Sep 13, 2017
    paperquagsire, JustinTR and Altissimo like this.
  7. Brucolac

    Brucolac

    Joined:
    Jun 5, 2017
    Messages:
    20
    If one gives me data, i'll do a statistical test.
  8. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    Wow, I didn't know people take my response so seriously, when I didn't even mean those stuff in the first place.

    To GG Unit:
    I dont know how you derive that is the only conclusion to what I said. Because it is very simple that there is evidence that the team generated (e.g. multiple different sets of pokemon, and the pokemon that will pop up), is based on some parameters of your battle.
    For instance, (at least based on what I heard) you never encounter legendary until you get over 40, you rarely get full ev pokemon until over 30, but you will almost get a lot more cheesey sets pre-30.

    All those are pointing to the obvious conclusion that the programmers of pokemon did think about the team formation and the overall difficulties they want to make to the players. If so, then this is not totally random.
    So I don't know exactly what parameters they take into account, but I do notice that overall the rules above is pretty much true.
    But feel free to disagree since maybe someone did met a legendary pre-40, who knew?

    Trainer sequence does change based on my team though, how that translate to AI cheatin?

    If you have done a lot of coding, you will understand that most successful game programmers when they have to build something with peusdo random levels that has good game balance, you wouldn't do completely random, because 1) it is almost impossible to make 2) you can't allow players to have some expectations 3) you can't add custom stuff to it to guide into direction you want the tree to give the right experinece to a large amount of players with variety however

    We can first talk about trainer sequences:
    I can clearly say that based on my many teams in the last beating the tree, in the process (I didn't document this) I noticed a general pattern.
    A lot of time when I use team A with pokemon and with those movesets, significantly more sightseer (almost all the time) will come up early post 20 to 30, or another team I will see more breeder post 20 to 30 without much sightseers, even another team I will see the old grandpa and grandma will show up more, it is interesting to see those patterns to show up for a particular team.

    I will use one team, beat tree 20+ times, and almost consistently seeing patterns like this.

    Let's talk about pokemon moveset sequences:
    This is a bit hard to notice since there are so many pokemon I fought in the tree, but I do noticed that especially post 20, if you have a team weak to rock, you will get a lot of pokemon that is weak to your team in typing but has movesets that super effect you all the time.
    And that goes hand in hand with the trainer sequence I mentioned above, as based on serebii, you get a trainer catelog and you see all the pokemon they could have, based on the pokemon team and their typing, you will get different distribution of trainers because of their pokemon typing and moveset. Also, I can't give you a specific example or even let you reproduce specific scenario, but this is what I noticed.

    Maybe I interpret your responses wrong, but I'm simply claiming that the game is not completely random, and react based on some parameters that includes your team's pokemon.

    I'm not here to say AI cheats, in fact, I prefer to take this as a challenge, because if you can predict AI's general movement, you can counter it easily. I prefer AI to not be completely random, that will kill the fun for team builders like me.

    I do respect people like you who spent a lot of time thinking about your team. Without people like you, there is no me. I simply can't fathom the vast amount of time you have to sink into this to get a team that works.

    However I am open to critique, as I'm wrong a lot of times.

    Yes, I don't know if you were on discord channel, but I did talked with GG Unit and learned and understand how to use the team a bit better now. I realized I was still learning, and probably made some very arrogant statements since I was frustrated. So I would apologize to those people who actually spent the time to build the team and can workaround threats or more like cheesy strategy like this when I didn't know what to do.

    In my comments, sometimes I was half joking. As it is mostly speculation.

    I do appreciate all the hard work that those people in leaderboard has sink their time into, as I am almost doing the same thing as well, thinking very hard. But I'm still at the novice level so that's why I can't get that far.

    So I don't know the full extend of that the durant team can do, as I prob didn't spend as much time as you all, but I think we are still mixing two things together.
    One is that i believe I was not saying that the team shouldn't get the glory it deserved if it was part of the battle tree streak, but I'm saying that the leaderboard phrased it in such a way that it seems like those people on the leaderboard use only one team to get to those streaks.

    And yes, I don't have any evidence to say that each person might be using multiple teams to achieve that, as they probably don't even track of those themselves, but the fact that people do switch teams, and battle tree allows it, makes it really hard for them to really claim that unless a better mechanism to proof is in place.

    I'm also not asking that, since I recognize the hard work. I can't say for others, but all I'm saying is, if you use multiple teams, please list out all teams you used in the streak. If you really use only one mono team? Awesome! Say it out with proud but you do realize you need further proof if you want people to be convinced, if you don't care then you know, don't care.

    However I do believe people here in this thread care about those stuff, that's why the leaderboard was created.

    If the teams on the leaderboard lack credibility, then I wouldn't try to find rental team, or try to recreate the teams on the leaderboard, since it means very little if that is the truth.

    The issue I have with the leaderboard, let me say it in another way, is that there is a probability (doesn't matter how small) that the teams on there is merely the last team they use before they lose. So in 90+ streak, that team posted might just be contributing to the last 10 battles.
    I'm not claiming that is true, but the mere possibility, makes it hard for people like me, to really take this thread seriously. And I don't want that to happen.

    That's why I'm still here, that's why I still willing to share my results and foster a good community to break the battle tree together.


    I hope that explains my stands a bit better.

    Also thanks to all the veterens here. Worldie posted some cool battle tree guides.
    http://www.smogon.com/forums/threads/battle-tree-mechanics-and-guide.3613222/

    That looks really comprehensive, looks like a must read for new comers to overcome their weaknesses in the tree.

    MOD EDIT: Please don't double post.
    Last edited by a moderator: Sep 16, 2017 at 6:39 AM
  9. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    I am sorry to have to burst the bubble here, but there's a few parts you are being mistaken, which I can forgive as you don't have as much practice with the facility just yet.

    Firstly, not all trainers can appear at every rank. Every 10 rounds have fixed trainers that can appear in them.
    In specifics, trainers with sets 1-2 appear only up to 20, trainers with sets 2-3 appear upwards to 40, trainers which run sets 3-4 only without legendaryes appear from 30 to 50, several trainers carrying legendaryes start appearing at 40, and there's a large bunch (like, 70+ trainers) that only appear post legend rank, and that generally sport plenty of Legendaryes, including mega versions, and even mono-legendary teams.

    As I said in another reply, it is a misconception that "if your team has several pokes weak to something, you will often have AI enemies that have supereffective".

    Expecially Pokes of sets 3, but in general most pokes that appear past Legend are optimized. This mean most of the sets that are not defensive will have perfect or near perfect coverage to hit everything for at least normal damage.
    Set 3s specifically are in general 4 attacks sets which run perfect coverage.

    Specific coverages are also extremely common: Earthquake, Blizzard, Focus Miss Blast, Dazzling Gleam, Rock Slide, Thunder, Icy Wind to mention some coverage moves that you see every other set 3-4.

    If you are running a comp with stacked weaknesses, once you hit battle 30-40, you will start being destroyed because coverage moves are very common and the sets actually are optimized rather than completely random Spatk Garchomp. (aka Garchomp-2)

    It is *definitely possible* that the Pokemon you are carrying or the moves they have are included in some form in the random generator. But if you think of it a little, the amount of combinations is so insanely huge, probably over the billion, that it is absolutely not realistic to even think that someone coded a way to tell the AI to pick up sets that are able to hit you against your stacked weaknesses.
    Codraroll likes this.
  10. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    I agree with what you said. Thanks for clarifying that, I do recognized that trainers and sets do seems to be mostly fixed to a specific level.

    However, I think I didn't state that
    "if your team has several pokes weak to something, you will often have AI enemies that have supereffective"

    I think I was merely pointing out the fact that the AI will react according to your team's typing.
    Also I didn't say it will be "often", as this is a non issue if you are a seasoned team builder, your team is supposed to be able to react and counter those typing from those teams anyway, so i don't see that as an issue. And if it is often, then you know what to think when you are building a team. And that is a good thing.

    And I agree with what you said that statistically speaking, since your pokemon team has to be able to counter a lot of different movesets and threats in order to progress, you will inevitable encounter super effective moveset no matter what.

    I will slightly disagree with your last statement though, the fact that there is a battle tree trainer catalog, shows that the developers also don't do a full random team generator. It is picking from a specific pre-set that they programmed somewhere in the game so the team comes up consistently with specific strength and weaknesses.

    I could imagine they have get a lot of inputs from competitives players in the past to build the teams so it gives players a challenge. So no, I don't think the probability is low, because they do try to react a bit (not saying it is immediate) but enough times so you feel the pressure if you team is subpar.
  11. Altissimo

    Altissimo

    Joined:
    Feb 19, 2015
    Messages:
    329
    HOW THOUGH? You aren't providing any sort of evidence for this, you're just saying the vague "the game responds to your typing". Please tell us how you think this happens in a very descriptive and in-depth way.
    Codraroll likes this.
  12. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    Essentially, the issue is that the sensation a lot of people have of only starting to be constantly supereffectived against once you reach battle 30 (where set 3 starts to appear often) leads to think that the AI is specifically trying to counter you with Pokemon that can beat you, when in reality, you are simply at that point running in actually optimized sets that generally sport perfect or near perfect coverage.
  13. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    Yea, perhaps this is all pointless discussion.

    Let me state this again
    1. I don't think AI is "cheating" even if they adjusted the accuracy of specific cheesy status move or items (e.g. quick claw, hypnosis), which is prob not true.
    2. I don't think AI is "cheating" because you get a lot of super effective moves (if your team sucks, you get that more often pre-10 even)
    3. I don't think AI is "cheating" because I can't get a leaderboard team to post-50, it is very well my lack of understanding how battle tree works, and how pokemon battle system works.

    All the above is something I can learn and take control.


    As for the claim that I believe team is reacting based on your team, I still think that the order in which the trainer appears, in specific levels (yes even if they are more likely to appear in specific levels) can be and should be observed based on the team you have.

    I don't know what is the model they are using, but I would imagine it would be something like this:
    1. They look at your team's lead typing
    2. They look at your past battle tree's moveset used

    I don't know the weighting, as it could be less than 20% or 50%, I have no clue.
    But some general patterns seems to be observed based on my sample of experience when I play the tree.

    In other words, what I'm saying is
    1. I'm not claiming that because AI react to your team quicker, it is cheating. it simply means if you have a bad team, it will ease your misery quicker
    2. I do think it is true, that the pokemon pool that battle tree can extract from is relatively small compare to the vast amount of possible teams, but very highly trained and seasons and smart combo that also post a lot of interesting challenge to different teams, therefore it is inevitable that your team will meet super effective and counter teams that can easily destroy you.

    I haven't changed my stands too much.

    And like I said, I could be wrong.
    Last edited: Sep 13, 2017
  14. Altissimo

    Altissimo

    Joined:
    Feb 19, 2015
    Messages:
    329
    This still does not answer my question because that does not explain how it affects what trainers or Pokémon are "drawn" by this.
    -Does the generation tend towards Trainers/Pokémon with super effective attacks or types against yours?
    -Does the generation tend towards Trainers/Pokémon weak to yours?
    -Does the generation tend towards fast Pokémon when you use slow ones, or high-attack Pokémon when yours have low Defense, etc.?
    -Does the generation tend towards Pokémon that won't take much damage from your attacks?

    If there is no discernible pattern, then it literally is no different from random generation whether certain Trainer types appear more often than others.
  15. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    I might have worded my reply wrong.

    I'm not saying you are saying that the AI is cheating.

    I've merely stated why people tend to have that thought when hitting battles 20-30+.

    As far as the rest, Altissimo's reply is on point: while occasionally it is possible to run into counters more often, the opposite also happens regularly (if you pass by Discord often enough, you can catch me alternating between complaining of Fire trainers, and memeing how stupidly easy per me Raz and his friends are anytime I get them in a sequence while they give tons of issues to a lot of other comps).

    Definitely possible that the random generation takes in account your team comp. Pretty much as possible as taking into account literally anything currently in the 3DS RAM.

    But there's no proof or consistence in showing that the generation actually has any direct correlation to your team either, since there's no consistence in what trainers you fight.
  16. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    I think if I understand you correctly, you are saying
    If there is no evidence that the code is reacting or using your team as input with some reproducible team, then perhaps the results doesn't matter and can be treated like a completely random RNG.

    A very weird logic, as you are not really disagreeing with me, but more like saying it might not matter. Let me know if I misunderstood you.

    Also I didn't mentioned anything about speed, or even attack or special attack, as I didn't think they look at your pokemon stats I think, but I could be wrong.

    Let's make sure we are speaking on the same page
    1. When we say generation, we are really saying that based on existing teams and trainers they have (we have the battle tree catalog), which trainers and teams should appear first, and which should appear later, and what would be the set they use to face your team. I don't think it is a completely random generation where you can get a super improbable pokemon like misno. as there are rules and restrictions on that.
    2. When you say random generation, I don't think you mean that there could be equally possible to get all pokemon in the pokedex to appear in battle tree, but you are saying that the typing and moveset that those pokemon will contain, is completely random against your team.


    I will try to answer your specific sub-question
    -Does the generation tend towards Trainers/Pokémon with super effective attacks or types against yours?

    I think so. Basically if they can find pokemon (of course within limits based on the level you are in) the team can do more damage to your pokemon lead, I think they will do that.

    -Does the generation tend towards Trainers/Pokémon weak to yours?

    I don't know, but why?

    -Does the generation tend towards fast Pokémon when you use slow ones, or high-attack Pokémon when yours have low Defense, etc.?

    I don't know, seems unlikely

    -Does the generation tend towards Pokémon that won't take much damage from your attacks?

    I don't know, seems unlikely
  17. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    You are slowly getting on same page.

    This is the catch: once you reach a certain number of battle, nearly every non staller Pokemon the AI has will have 1 or more supereffective move against 2 or more of your team composition.

    So even implying this was true, it becomes irrelevant, because out of say, 20 Pokemon a certain AI trainer has access to, 12-3 will have 1-3 supereffective moves against your team, with another other 5 being stallers. (that is, outside of specialists, who may or may not be easy for your composition, or completely painful)
  18. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    Yea, I recognized that most people only complain when they lose the battle (me included), and it is possible that if they were given the chance to continue from there, that battle might seems to be occuring rarely.

    However no proof doesn't mean it is not there, in my bias, and thinking in their shoes, if I'm a game developer, as I stated earlier, it wouldn't make sense for me to have this a completely random battle tree.

    There has to be specific rules, and things that could make each battle unique and post challenge to your team.

    Otherwise, I think based on the complete RNG logic, all I need to do is to keep trying with the same team, or any random team (doesn't matter if it is UU/OU/ or less) perhaps 1000+ more times I will eventually get to the top right?

    Then why do we need leaderboard then?
  19. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    Thing is, it DOES make sense to have a completely random battle tree.

    The key of Battle facilities *IS* the randomness.

    There's a insane amount of things that are randomly generated: The trainer you face, which Pokemon he will have, which sets of said Pokemon, which abilities the Pokemon have, and there's plenty of further RNG added via the sets themselves with 1HKO, inaccurate moves, Evasion/double team, and the list goes on.

    Like what I wrote in my WIP guide, this generation it seems they went one step further by also adding a chance for the AI to do completely unpredictable and non sense actions like double/triple protecting, picking ineffective moves instead of supereffective 1hkos, swapping for no reason, and several others.

    The whole thing that makes the Battle Facilities challenging is the fact that you never face the same thing, and even the same trainer can be completely different between 2 battles.

    The randomness IS the base of the Facility, and why you need failsafe compositions to pull the 100+ long streaks.
  20. Altissimo

    Altissimo

    Joined:
    Feb 19, 2015
    Messages:
    329
    If the AI is pulling Pokémon that seem to pose a threat to yours, how is that NOT an attempt at counterteaming?
  21. Pencilcheck

    Pencilcheck

    Joined:
    Dec 20, 2016
    Messages:
    27
    Perhaps I worded wrong then, yes, there has to be some randomness in the teams in battle tree, but not like completely random. Perhaps I misunderstood what others are saying as well then they say completely random.

    I don't think I changed my stance for that at all, perhaps I just didn't expressed that well?

    MOD EDIT: Please don't double post.
    Last edited by a moderator: Sep 16, 2017 at 6:38 AM
  22. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    Perfect wording is definitely opinable, as "random" in IT does not exist to begin with.

    To use a more formal wording, the potential data used for generation of the opponent you will be facing has never proved to be consistently attempting to create compositions that are strong against yours.

    Expecially considering that, on top of what I said above, "strong against" is a extremely volatile term when it comes to Pokemon. Just because your team is slow, it does not mean that it's weak vs fast teams. In same way, just because you have 3 Pokemon weak to Flying, that does not mean your composition fears fliers.

    If you think about it for a moment, you can realize that simply having supereffectiveness available is not a definition of "countering", expecially in the light of most set 3-4s and Legendary sets being optimized to begin with, due to how volatile team compositions and synergy can be.
  23. Brucolac

    Brucolac

    Joined:
    Jun 5, 2017
    Messages:
    20
    For the issue of wording, i would suggest to define any attempt (may it be there or not) from the AI to gain an advantage by chosing not completely random Pokemon as "countering".
  24. Worldie

    Worldie

    Joined:
    Apr 6, 2017
    Messages:
    166
    Which is the issue: what is the definition of "attempting to gain an advantage" in Pokemon battles? Outside of actually meme/trash sets like Florges-2, it is a definition that doesn't make sense, as every Pokemon has weak and strong points.

    A Fire Pokemon is not necessarly weak to a Water Pokemon if it has access to coverage and is faster or has priority. A slow Pokemon is not necessarly weak to a Fast one. A squishy Pokemon does not necessarly fear powerful hits. A pokemon with terrifyingly bad offensive stats is not necessarly not capable of 1HKOing.

    I can continue for years on this subject.

    There's no realistic way to even *think* to define "gaining an advantage" in Pokemon, let alone code it.
  25. Brucolac

    Brucolac

    Joined:
    Jun 5, 2017
    Messages:
    20
    The end of the sentence is particularly important:
    "...to gain an advantage by chosing not completely random Pokemon."
    No matter what thoughts of the game developer hide behind this not-completely-randomness (if existing), i would call this "countering".

Users Viewing Thread (Users: 2, Guests: 11)

  1. NZtechfreak,
  2. d4rk