Results 1 to 10 of 10

Thread: Welp. I'm off again fellas.

  1. #1
    Basic Member
    Join Date
    Jun 2013
    Posts
    274

    Welp. I'm off again fellas.

    Come back to the forums and find that machine learning is taking over bot scripts as well. XD
    I don't think I can keep up anymore. It's both fascinating and depressing.

    Best of luck.

  2. #2
    Basic Member
    Join Date
    Dec 2016
    Posts
    46
    Do not be discouraged, machine learning is not everything. Bot scripts still need the traditional approach. Even on the day of self-learning, it still needed to combine the two approaches.

    I suggest you to read this post:https://dev.dota2.com/showthread.php?t=283655
    Last edited by adamqqq; 11-17-2017 at 12:28 AM.

  3. #3
    Basic Member
    Join Date
    Jun 2013
    Posts
    274
    I sure hope you're right. Like custom games, Bot scripting became another thing to do in Dota 2. I'd hate for it to become pointless because machine learning makes better bots than we ever could.
    I'll hate it even more if I need to spend a few years catching up on machine learning just to be relevant in a pastime. XD

  4. #4
    Basic Member
    Join Date
    Dec 2016
    Posts
    180
    I have same feelings too. Decision making makes me annoyed to think about gameplay of entities in the map. its hard and actually we are not valve developers .

  5. #5
    Basic Member
    Join Date
    Jun 2013
    Posts
    274
    I'm back to scripting again, but some of my issues still exist. The 3 key one's being:

    • If there were to be bot vs bot events, fully Lua bots won't stand a chance again learning bots.
    • Even if everyone were to have the ability to make learning bots, it seems pointless in the long run since only 1 is really needed.
    • Said learning bot would be the most used and all other standard bot scripts would be left to collect dust.


    Human-made bot scripts need to be manually updated with all the situations, conditions, reactions, etc. hard coded.
    Learning bots will, by their very purpose, find ways to do everything the best way possible on their own or with added replay data.
    To be competitive, one needs to be added to while the other needs to be taken away from. Very different approaches.

    Multiple learning bots seems pointless, once again, due to their purpose. They will eventually reach an equilibrium. In that case, isn't is just more logical to pile all the play data into 1?

    The last point I'm split on from a player and scripting perspective. As a player, it's neat knowing there's a learning bot that's guaranteed to have a different performance every time due to the influx of players or replays adding to its learning. As a scripter, it sucks knowing that your script that you put your own personality and play style into doesn't really have a place anymore.
    It may also sort of suck as a player too because I like checking out different scripts to see what they do differently.

    This is mostly assumption, of course, since we don't have any learning bots out in the wild yet. But it definitely feels like this is how it's going to go.
    tmw you realize machines are even taking over your leisure activities. X(

  6. #6
    Basic Member
    Join Date
    Mar 2012
    Posts
    2,017
    Quote Originally Posted by Cornbane View Post
    • Said learning bot would be the most used and all other standard bot scripts would be left to collect dust.
    But if another bot kicks their ass and the losing bot would analyze all lost matches, wouldn't he improve as well? Point is, there will never be a constantly perfect bot, just like the same teams don't win each DOTA tourney. They constantly rotate because each team improves and "steals" a few tactics from other teams or the master frog changes the game and everyone has to relearn it
    Explanations on the normal, high and very high brackets in replays: here, here & here
    Why maphacks won't work in D2: here

  7. #7
    Basic Member
    Join Date
    Jun 2013
    Posts
    274
    For people I can agree. Because we have a limited cranial capacity. We have hesitation under pressure and delayed reaction time. We also tend to repeat the same mistakes, no matter how good we get.
    Machines have virtually unlimited storage space and react in mere single digit milliseconds. And let's not forget that a major advertising point behind them is that they can learn at levels that are beyond even our wildest comprehension.

    Anyhow. As for the point you referenced:

    For Lua bots, they'll be ignored because they're mere Lua bots. There's a learning bot over here that's more advanced! Though they may be forced down to Lua level after failing to win a few hundred times.

    For learning bots, you know how people work: They'll see the one with most downloads and highest rating and go with it as well because those are assumed signs it's the best. Even though using the other one's will help them improve or they may already be just as good as the top one.
    Look at the Torte De Lini hero guides. Right in this game we can see how number attract more numbers. XD

    What I'm still wondering is how will these learning bots work on local play. If you can't connect to the server, you can't use the bots. Unless you host it locally, which may cause problem for the one hosting it if it requires a lot of processing power.
    I suppose these will be answered if ChrisC has the heart to return after all these 7.07 fixes.

  8. #8
    Basic Member
    Join Date
    Dec 2016
    Posts
    731
    Learning bots can work on local play in two ways:

    1) Connect to a remote server
    2) Use a trained policy

  9. #9
    Basic Member
    Join Date
    Dec 2016
    Posts
    731
    Look, don't get discouraged. Lua coding is not going away. All bots for Dota2 still need a Lua backend to implement whatever they decide. Learning is a very large and immature research endeavor currently. You think people like DeepMind and OpenAI know what they are doing, and they do have sound logic, but they are exploring new concept just as much as you and I can.

    IMHO, the next decade of work in technology will focus primarily on problem decomposition and layering. So no, you aren't out of a job, but do pivot somewhat to focus on how to break down massive problems into smaller inter-connected problems. Also, don't focus on the problem, focus on the solution.

  10. #10
    Basic Member
    Join Date
    Jun 2013
    Posts
    274
    I think this topic got lead astray somewhere. XD

    Anyhow, I had an epiphany. For bot vs bot events, just have different pools for each script type. i.e. One for fully Lua bots and one for ML bots.
    A solution so obvious that it took several days to realize.

Posting Permissions

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