draftbots and afk autopicks will more aggressivly take rares #5397
Conversation
two draft quality of life improvements for situations when real players quit or go afk. 1. made the draftbot lean on rarity for card ratings. this helps make the draftbots behave and not just pass bombs to the human players. 2. changed the draft autopick that happens when the timer runs out and the player hasn't selected anything. I changed it to pick the last card in the pack, which should be the rarest, so basically it raredrafts. again making it so AFK players aren't just feeding real live human players the best cards.
|
Some remarks about score -- it's not only draft. Score system uses for AI game's decisions too (it's checks all possible actions, plays it as simulation, calcs new table score and selects action with max possible score). That's why that's consts must be selected with caution. |
ok, well using rarity for threat assessment isn't a bad idea either. how serious is the ai development for this project? |
|
I have no idea why these 5 tests are failing: testSummonersEggImprint(org.mage.test.cards.facedown.SummonersEggTest): (Hand) Card counts for card Sejiri Merfolk for PlayerA are not equal expected:<1> but was:<0> none of these tests even look like they're using AI in them. anyone have any hints about that? |
It's not at development now. AI can't play combos. AI can't use some hard cards/abilities too. About tests failures: I think it's about new rarity coefficient in score calc. You can try to change it from 80 to 40 and etc to test. I'll look later on it too. P.S. Test engine uses AI for card playing. |
ok I see whats happening, the tests use the rating to figure out which card to imprint. some options, I could move that rarity rating checker just specifically to the drafting part. |
|
Nope, it's a good idea for AI too (not only draft). Need more research. It's a test engine bugs, not yours (at least for one of the test). As example: failed
|
…arity multicolor cards are the payoffs for playing those colors so they should be better reduced the effect of rarity increasing the rating
addtarget was being ignored and AI was being used to determine a target. the spell has a target so I think this is not a bug with add target but rather the test itself
this test uses ai to determine which card to imprint, with my other changes to AI, the rarity of the cards changed the relative rating of these two cards. I changed the merfolk to be a common merfolk so that part of the card rating shouldn't affect it
|
ok, so my changes fixed the tests. the summoner's egg test shouldn't rely on AI at all, it should really hardcode to specify which card to imprint. I couldn't figure out how to do that so I just changed the card name to a common. summoner's egg isn't a targeting spell and the addTarget function didn't seem to allow me to set it. imo, there are probably lots of tests that implicitly use AI and that should be avoided. I have a few more ideas for updates to the AI, and I guess I'll just have to update the tests that they disrupt as I find them. |
two draft quality of life improvements for situations when real players
quit or go afk.
the draftbots behave and not just pass bombs to the human players.
the player hasn't selected anything. I changed it to pick the last card
in the pack, which should be the rarest, so basically it raredrafts.
again making it so AFK players aren't just feeding real live human
players the best cards.