Funnies thing is that it lends credence to all the conspiracy people that claim draw and RNG are rigged.
They clearly have the code to do it, and they are showing it off. Numbers still don’t back up the conspiracies, tho.
Morris Code
function cards(gamestate) {
return options(gamestate.mana, gamestate.playerHero)
}
function options(mana, hero) {
return hero.cards(mana)
}
Class Mage: Hero {
function cards(int) {
switch(int) :
case 1:
return (randomSelect(Arcane Blast, Arcane Missiles, Breath of Sindragossa), randomSelect(Arcane Artificer, Babbling Book, Kabal Lackey), randomeSelect(Ray of Frost, Shooting Star, Tome of Intlellect))
}
}
2 Likes
Zephrys cards are not restricted to your class and it is a bit more complicated that just looking at how much mana you have.
snaps fingers
Invisible Ink!
Agreed. I wanted to imply that this is not some next-level neural network calculating chess moves. “a bit more complicated” is likely technically correct. The best kind of correct.
After coming back, I also want to point out that I correctly indented those code blocks, and this middle ages markdown editor flattened everything. I cry foul.
1 Like
Look i m not saying its easy. But just seeing how complicated, things have evolved to be in the digital world zephrys looks so simple to be coded.
Just take a look at any game around you look at the graphics, at the bots and how the AI enemies work these days.
Most likely a complex CASE statement.
I believe Ayala mentioned on one of the recent Omnistone episodes he was on that they’ve actually sunk hundreds of man hours across multiple people and literally thousands of lines of code behind Zephrys.
1 Like
They could of used SWITCH statements as well. It seems like a good way to chunk the data at certain points of the logic.
2 Likes
Tbh I don’t think they did use machine learning (like with alpha zero) for zephrys,it is a kinda complicated way of doing things which for hs probably is not needed.
I think they made it exactly like they made the chess programs that don’t use machine learning. Search trees in combination with an evaluation of the boardstate. It would be relatively simple since it would be only 1 or 2 play deep,maybe 3 at the most for next turn but it most likely wont even take into account your opponents possible answers.
I am curious how they did this,could maybe blizzard say if they used machine learning or a search tree in combination with board evaluation?
I think you are the most correct so far haha. This is not the best use-case for machine learning. There isn’t really a clear fitness measure and its effectively a static amount of situations(albeit alot of them).
Could you do an initial switch statement to determine the “biggest threat” so to speak. Then reference global object arrays when going into the if and else statements?
Heres a quick repl I wrote to explain what I mean a little.
h`ttps://repl.it/@Dillon_temple/HS
Im still learning too, so this is a good thought exercise haha.
1 Like
Right?? It’s the first time I’ve stepped outside the boundaries of my coding class’ assignments to look at programming from a recreational angle, and it’s really fun. I’ve just started learning about switch statements and they are still super fuzzy to me, so this is good.
I wonder if there is a codewars.com kata problem that attempts to do something similar to this…
1 Like
When you’re ready to take it to the next level, start thinking about how different languages would tackle this problem. An object-oriented language’s implementation (probably similar to my pseudo-code above) would look different than a functional language’s implementation (which would likely use closures/curried functions). And strongly typed languages would give you access to things like Reflection and property…properties, which are cool but usually too obtuse.
Not to call anybody out, but I seriously doubt that the card is coded based on learning algorithms. That would imply that the card could behave differently at different times for different people as it “learns”, and that’s the kind of thing which is usually called out as “unfair”. I imagine that the offered cards are ultimately the returned values/pointers of a deterministic function.
3 Likes
No. Once you finished training it doesn’t change.
Correct. But why are you sure
- They’re using learning?
- They’ve finished?
I’m not sure they are using machine learning, but I can see from your post that there are some misunderstandings concerning machine learning.
What?
Okay, I’m paid to write self-improving code, and teach programming. I’m not saying I’m a next-level expert (though I know several who’ve had plenty to teach me), but if you’re calling me out for ignorance because I didn’t provide a textbook or whatever you were expecting as an anti to something you don’t even believe, then cool, bro.
If you have credentials and something interesting/constructive to say, I’m always fascinated. If you’re just another internet nerd throwing out "actually, " because you derive your self-worth from putting people down, it’s too early in the day and I have too much work to do, so checking out.
Thanks, I appreciate this post! I’ll keep the advice and circle back around to it when I am ready to branch out. Currently they have us learning a lot of vanilla javascript and react libraries for web development, but I don’t want to stop there once I feel confident with those. That’s a ways off though; what I am chewing on now is still pretty fun and daunting 
1 Like
Tough projects like that are the way to go, IMO. Even after all these years, business organizations want you to try tutorial after tutorial, which are good for learning how to set up a framework, but at the end of the day, the code you write to implement a basic API in LAMP and MEAN (or whatever the kids are using these days) is not very different.
Find a framework you like and challenge yourself, NEVER assume that you know everything, and yeah, if you have any questions, feel free to hit me, or whomever, up. Everybody has something to teach and something to learn in this business.
1 Like