Page 7 of 8 FirstFirst ... 5 6 7 8 LastLast
Results 61 to 70 of 73

Thread: [Feedback] Bot Item Build Changes (still WIP)

  1. #61
    Basic Member
    Join Date
    Feb 2013
    Posts
    222
    It's especially bad with Viper though because he does more damage the more damage they've taken through Nethertoxin which they simply don't realize.

    Personally I think Crystal Maiden could use a looking over considering that people seem to think she may be more effective if she only puts one point in her aura in favor of maxing frostbite and nova first. Not to mention that she's STILL got phase boots in her build but any of this should be kept to its own unique thread for each bot but we should only deal with 2-3 at a time.

  2. #62
    Basic Member
    Join Date
    Dec 2011
    Posts
    1,903
    Right so I got delayed some more experimenting, but here it is. Scroll down for the file in attachments.

    There’s loads of changes in this file, to name the obvious ones quickly:
    • Optimized starting items to optimally use gold (taking into account likeliness of supports/gankers buying wards/courier)

    • Improved skill builds across all heroes, most notably making CM 4/4/0/0 into stats (mostly because I prefer this build) and making WD not skill his heal until the very last (because he just pointlessly burns his mana with it)

    • Improved item builds across all heroes, most notably turning Tiny into a carry, of course many item choices are debatable, but I hope that my changes aim to improve bot builds in all situations.
      That means the items they buy are usually always good items and early items are easy to buy (for example, rushing a Heart on CK after armlet is fairly common, but due to the difficulty of getting 3k gold, I opted to have him get Halberd first, as it consists of much small parts than Heart)
      Furthermore, I also removed Necronomicon from any bots build. The AI for using the item is terrible, they barely ever use it. And even when the Necronomicon units just stand about aimlessly, supposedly they had their own AI since 2 weeks ago but I don’t notice it. (And even if they did I still don’t think it’d be a good item as the bots use it about once every 5 minutes if you’re lucky)

    • Adjusted laning info on many heroes, most notably reducing provide babysit on carries across the board.
      While it’s true that a number of them don’t need much help in the lane, there are rare cases where due to providing babysit it’ll match up 2 carries for one lane and 2 supports for the other.
      The AI should more so use this provide babysit as some sort of indicator of ‘Ah right, this hero doesn’t need that much help in the lane, so it’s okay if I send a fairly weak laning support like Tidehunter with him’ rather than a cue to send 2 carries into the lane. Since this isn’t the case, I just changed it as a temporary solution



    Now for the bigger part, namely the role changes, there’s actually quite a lot of thought and experimenting that I put into this, sadly I don’t think I can perfect it as the AI is a bit flawed at times, sometimes being completely unwilling to pick certain heroes for no obvious reason, often times changing the laning info to make them more likely to pick that hero merely results in a different hero becoming hated by the AI.

    So first off I’m just going to report some observations from testing the picks
    I noted down the picks of 30 games and got the following result:
    Code:
    Hard Carry	60
    Semi Carry	55
    Ganker		72
    Stun Support	73
    Pure Support	9
    Push Support	31
    Now the first thing to note here is that the bots will -always- pick 1 Hard Carry. Secondly, they’re pretty consistent at picking an average of 1 Semi Carry, 1 Ganker, 1 Stun Support, and then the last pick is spread across the Ganker and Support roles, mostly in favor of Push Support.

    Due to the Pure Supports being picked up so little I opted to make the heroes marked as Pure Support also be considered as either Push or Stun Support.

    Furthermore, as there will never be 2 hard carries on the same team, I somewhat redefined hard carry. I defined hard carry as a hero ‘whom you’re not going to want another one of’.
    Simply put, picking another hero marked as hard carry will probably weaken you to such a degree in the early or mid game, or put so much emphasis on the late game, that your team will suffer.
    As such, some heroes had their Hard Carry status revoked, like Razor, Viper and Juggernaut, as I don’t believe your team suffers from having them together with a hero like say Drow Ranger.
    Additionally some other heroes were moved up to Hard Carry status, like Chaos Knight, Skeleton King and Dragon Knight.

    Scroll further down to see the full chart.
    Also, before someone raises questions about CM as Stun Support, I did that mainly because I didn't want to have more Push Supports than Stun Supports, and she had to be marked as something other than Pure Support.
    Furthermore, WR is marked as semi-carry because I like playing her as such and there's already enough heroes with Mek.


    Finally, as to the reason why I uploaded this (Viper’s OPness)
    I did some testing beforehand, where I focused more on his performance. I watched about a dozen games with him in it, and basically what I got from watching him is that he’s overpowered in the sense that he -always- perform well. I only caught 2 games of him where he was sitting around 250-300 gpm, apart from that he’d always fluctuate between 400 and 600 gpm, usually from about 20-30m+ onward he’d stabilize near the other carries though.
    He had a pretty high win rate, I believe mostly due to the fact of how well he secured victory in the mid game. It makes sure that the Hard Carry on his team has an easier time getting his late game up than the enemy’s Hard Carry. (Among the games he lost, this one was by far the most amusing, that Jakiro…)
    Regardless, I didn’t exactly feel he broke games. He’s a strong presence in every game he’s in certainly which makes him much stronger than most bots, but I didn't feel he was as broken as people make him out to be. There were only 1 or 2 games where he completely went out of control, but for the same matter I had a game where a Sniper ended up with 750 gpm and his team ended the game 23 minutes in. (Viper was on the opposite team)
    Attached Images Attached Images
    Attached Files Attached Files
    Last edited by Arxos; 03-24-2013 at 11:19 AM.

  3. #63
    Basic Member
    Join Date
    Dec 2012
    Posts
    5
    Really nice Arxos I will definitely look at your file

    Also i've got another doubt,I found this:
    Code:
    "BotForceSelection"			"0"
    In Lina's section of the npc_heroes.txt and she's the only one that has that line. Anyone tested it yet?
    (I can't test it right now since I don't have Dota on this computer so...)

  4. #64
    Basic Member
    Join Date
    Dec 2011
    Posts
    1,903
    Quote Originally Posted by jptarrega View Post
    Really nice Arxos I will definitely look at your file

    Also i've got another doubt,I found this:
    Code:
    "BotForceSelection"			"0"
    In Lina's section of the npc_heroes.txt and she's the only one that has that line. Anyone tested it yet?
    (I can't test it right now since I don't have Dota on this computer so...)
    Haven't a clue.. by the name I imagined it'd force the AI to pick the hero if it's set to 1 (if you want to test a specific hero that'd be pretty convenient after all)
    However that doesn't appear to be the case, so I'm out of ideas.
    It may just be old code that wasn't removed.

  5. #65
    Basic Member
    Join Date
    Dec 2012
    Posts
    5
    Quote Originally Posted by Arxos View Post
    Haven't a clue.. by the name I imagined it'd force the AI to pick the hero if it's set to 1 (if you want to test a specific hero that'd be pretty convenient after all)
    However that doesn't appear to be the case, so I'm out of ideas.
    It may just be old code that wasn't removed.
    Maybe it just increases the chance of the bot being picked,or it really is just some old code that wasn't removed.
    It would have been really handy if it did force the AI to pick the hero :/

  6. #66
    Basic Member DarkLite's Avatar
    Join Date
    Dec 2011
    Location
    Patch Anticipation Station
    Posts
    3,919
    Arxos, you've got Lich and Omni tagged as Push Support and Lion/Lina as Ganker only rather than Stun Support - what's the reasoning behind that?
    Please help keep the forums clean by reporting rude or abusive posts!
    Got a problem? PM me or another moderator!
    Forum Rules


  7. #67
    Basic Member
    Join Date
    Dec 2011
    Posts
    1,903
    Quote Originally Posted by DarkLite View Post
    Arxos, you've got Lich and Omni tagged as Push Support and Lion/Lina as Ganker only rather than Stun Support - what's the reasoning behind that?
    I only used single roles for everything bar pure supports due to them being rarely picked up, though technically it's only necessary that hard carries and semi-carries don't have multiple roles so I suppose I could change that... The reason it's basically a requirement for hard and semi-carries is due to farm priority, the bots don't comprehend the fact that a support hero has carry potential, hence I've refrained from giving carries a second role under all circumstances.

    For Lich and Omni I tagged them as Push Support because of the low pick rate of Pure Support, and since neither of them have stuns Push Support is the only fitting one. That said, Omniknight is a pretty sick pusher in team format as he works similarly to Dazzle by pushing through raw healing power, so I don't think it's unfitting for him.

    In regards to Lina/Lion as only gankers, that is mainly because the bots really like gankers. Without them also being gankers, I can almost guarantee you'll see Zeus and Bane in every single game. For the reasons mentioned above I technically could mark them as stun support as well as it wouldn't affect performance greatly, however it would only serve to make Lion and Lina be picked more often, and since there are already a fairly low amount of gankers they really don't need their pick rate increased. This is also partly the reason why I kept to single roles for all non-pure support heroes, as say a hero that is marked as both ganker and stun support is plainly 2x as likely to be picked up as a hero marked solely as ganker.

    It's a bit awkward and flawed, but I feel it's the best way to get good and varied picks with the way the AI picks heroes as is.

    EDIT: To expand on the gankers thing, when I did my 30 games where I recorded their picks, which was before Lion/Omni were added (so only Zeus/Bane/Lina were gankers at the time), in the 30 games played Bane was picked up 26 times, Zeus 24 times and Lina 22 times. I'd love some more gankers but sadly the heroes most suited for it are all carries, and hence I will not mark them with another role. Arguably some heroes like VS and Witch Doctor can gank quite effectively as well, but that'd just be moving heroes from Stun Supports (which is picked up often) to Gankers (which is picked up often) so it'd just be moving the issue more so than solving it.
    Last edited by Arxos; 03-24-2013 at 04:35 PM.

  8. #68
    Basic Member
    Join Date
    Jan 2012
    Posts
    640
    Thank you for uploading your bots, Arxos. I will test them and try to improve mine.

    For now, the most effective "counter" my bots know for viper is Necrolyte and Sniper. BRB, spectating your bots =D

  9. #69
    Basic Member
    Join Date
    Jun 2012
    Location
    London, UK
    Posts
    678
    It think it's okay having multiple tags on the support heroes that can play multiple roles (with the single item build they have). I haven't noticed them overly picking certain heroes. As long as the first core items fit the support roles, like mekansms, they will do okay with less farm and still play that role. If they are allowed more chance to farm then they will progress to their later items. Heroes such as Lina, Lion, CM and WD excell at playing various support roles (and the bots seem to play these roles well).

    Lina is played really terribly when she's picked as a ganker, though. She dives too much and mostly feeds kills. She does better in the pure support or stun support role.

    A different solution is to remove some of the ganker tags. It works the same way. I took it off most of the supports.

    I agree with Omni being a push support but he's just a terrible support in the current dota meta (compared with those heroes, anyway) and I wouldn't be surprised if he doesn't function very well when played by a bot either. Since changing him to semi carry/hard carry (with push as well) - and specifying solo desire 1, farm 2 - I've seen him become the 10cs/min monster he can be and I can't really imagine him as anything else.
    Last edited by cinephile; 03-25-2013 at 12:57 AM.

  10. #70
    Basic Member
    Join Date
    Dec 2011
    Posts
    1,903
    Quote Originally Posted by cinephile View Post
    It think it's okay having multiple tags on the support heroes that can play multiple roles (with the single item build they have). I haven't noticed them overly picking certain heroes. As long as the first core items fit the support roles, like mekansms, they will do okay with less farm and still play that role. If they are allowed more chance to farm then they will progress to their later items. Heroes such as Lina, Lion, CM and WD excell at playing various support roles (and the bots seem to play these roles well).

    Lina is played really terribly when she's picked as a ganker, though. She dives too much and mostly feeds kills. She does better in the pure support or stun support role.

    A different solution is to remove some of the ganker tags. It works the same way. I took it off most of the supports.

    I agree with Omni being a push support but he's just a terrible support in the current dota meta (compared with those heroes, anyway) and I wouldn't be surprised if he doesn't function very well when played by a bot either. Since changing him to semi carry/hard carry (with push as well) - and specifying solo desire 1, farm 2 - I've seen him become the 10cs/min monster he can be and I can't really imagine him as anything else.
    If I were to remove the ganker role from Lina, I'd probably be more leaning towards removing the ganker role from all bots, as I'm not really interested in seeing Bane and Zeus in every single game... I'll consider doing some testing in regards to their picks if I do so.

    And well, to be fair I kind of consider Omnicarry a troll build. It's amusing to play, and sometimes it's effective, but in the end any half-decent proper carry will bash his skull in.
    And turning him into a carry build while marked as carry and push as well means you'll get in situations where he's picked as push and deferring all farm to his laning partner yet rushing a Midas, such situations are exactly why I moved to the single role setup.
    Omniknight is still an amazing support, the main reason Omniknight has fallen out of the meta is because he can't assist the current gank-heavy meta. He's mainly for countering ganks if anything, but with the quantity of disables going about it makes it hard for him to counter anything.
    Bots on the other hand play a push heavy meta, so I think he fits in just right.

    Quote Originally Posted by Karyoplasma View Post
    Thank you for uploading your bots, Arxos. I will test them and try to improve mine.

    For now, the most effective "counter" my bots know for viper is Necrolyte and Sniper. BRB, spectating your bots =D
    I usually find that Necrolyte does really well against Viper in mid. Death Prophet does as well if she decides to spam Swarm properly, but she often doesn't.
    Last edited by Arxos; 03-25-2013 at 03:04 AM.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •