We have to do some strenuous testing to figure it out.
Devs told us that arcane missiles goes in the pool if there is a 1 HP enemy minion. The rest we have to figure out ourselves.
1 Like
I know that if you have 4 or less cards, arcane intellect gets a bump in probability
It will usually cast that unless there’s a big threat (or a lot of threats) on the board or unless your health is very low
It does have a method behind it’s madness. To me it plays really intuitive plays most of the time. The most surprising play to me was frostbolt face 2 times in a row against handbuff paladin with a weapon equipped. I wouldn’t dare to expect that play, let alone twice in a row.
Khadgar MVP
Huh. Well, I am not the expert that you guys are, but the more i learn, the more this card seems like an answer in search of a problem.
It seems situational, and more like a “win more” card rather than being something reliable that improves the mage class as a whole.
I doubt that it’s weighted. It’s probably a binary in or out of the pool. So if it’s 4 cards for the trigger, it won’t ever cast it with 5+ in hand.
It’s the opposite of win more because it’s picking spells to solve problems, not extend a lead.
I’ll take your word on this, but I would far rather have a card that lets you choose from three spells to cast.
That’s not how machine learning in general works. Binary in or out happens in the end, in the final layer of the neural network, but the weighing and deciding happens using gradient descent algo which weighs the options with their benefits and seeks the one with highest value.
It’s pretty intuitive. If it misses sometimes, it’s because of matchup dependant stuff it can’t possibly account for. Well, at least not yet with our technology. For it work any differently would require constant unsupervised training, which would mean the game gets updated all the time.
Already got arcane intellect with 7 cards in hand.
So it’s not binary.
Well; we know this ain’t happening, lol.
1 Like
I can almost guarantee you there isn’t a machine learning algorithm at work here. The devs explained it already that each spell has a condition to add it to the pool.
It’s random after it gets put in.
It will never arcane missiles unless there’s a 1 health minion, then it might.
If there were clear-cut conditions, they would be easy for us humans to learn them in couple of games, because we would notice the patterns.
The outcomes are intuitive, but definitely not predictive, which means it’s not If-then-else conditional programming, but something else.
So far looks like machine learning, and not a bad one, considering the differences between matchups are huge
Well, I know I am always the downer, but this card seems terrible to me.
For the average player, they are not going to have any idea how to use this effectively, which seem to be Team5’s entire reason for cards like this.
Once again; a card that can be good, if you are Brian Kibler, but is mostly trash, because the player did not figure out the board conditions.
Meanwhile; Rogue kills you on turn 4 with no thought beyond playing on curve.
I stand by what i always say: Team 5 hates Mage, and wants it crippled, so too many players don’t use it.
There are, we just haven’t bothered testing for and writing them down yet.
The first is:
Arcane missiles REQUIRES a 1 health enemy minion.
RidiculousHat told us that one.
It’s meant to get better as we figure it out.
Sure, but that doesn’t tell you much about any of it. Even machine learning algo would emerge with that condition by itself because playing that vs targets with more health is never a positive value play so it wouldn’t ever be a winner compared to other options in the pool
If I read everything here correctly, no one beyond reliable Legend players are going anywhere with this card.
I despise the decisions being made at Team 5.
That card wasn’t even supposed to be a part of Rainbow Mage.
I think I have every reason to expect stuff that makes sense in upcoming expansions.
Right now we have other priorities to deal with, like infestation of demon hunters in all ranks.
It’s quite clear how it’s coded. We just can’t predict it because the requirements are too open-ended.
Mirror Image, Fireball, Frostbolt, Counterspell, Ice Barrier are always a possible choice in every scenario.
Just that makes the card hard to predict its outcome.
Polymorph is probably ‘only if enemy minion on board and above 1/1 stats’
Arcane Missiles is said above
Blizzard and Flamestrike are also ‘if even 1 enemy minion on board’ bec I think I had it cast Blizzard in one minion alone once.
And Arcane Intellect is just coded to not include it as a possible option if it will make you burn after, prob.
I’d be incredibly shocked if the programming wasn’t:
Cast a spell at random (equal weighting) from the pool of spells who’s conditions are currently met (prefers enemy targets).
With each spell with 1 condition that adds it to the pool.
So if your opponent has a wide and large board, and you have full health, plenty of cards and hand, no 1 HP enemies… the odds of flamestrike/blizzard get real high.
According to Orion up there, not even then xDD
But I can’t say I ever saw it make a stupid play. It was always something that made sense from some perspective, and I’m pretty sure I’m credible enough to say that as I play in higher ranks. I’m sorry, but to me it appears like some of you just don’t understand how useful it is because you lack knowledge about the game.
I don’t trust Orion but it would be a bug if it drew to burn.
That’s definitely not useful.
It could be, if you had 1 card left and it was Sif