Tempest world development
Moderators: timmermac, Blacksmith, KrazyTheFox, Agetian, friarsol, CCGHQ Admins
75 posts
• Page 3 of 5 • 1, 2, 3, 4, 5
Re: Tempest world development
by Seravy » 16 Sep 2017, 18:58
Any change to support the Warrior en-Kor, Angelic Protector, Worthy Cause infinite lifegain combo? if not, I won't bother building a deck that has it.
Ideally, the AI should be able to announce "I want to activate this 30000 times" as per the appropriate rules, if the player has to wait it out, they'll go mad
..but surprise infinite lifegain in a white weenie is too much fun to not include...
Ideally, the AI should be able to announce "I want to activate this 30000 times" as per the appropriate rules, if the player has to wait it out, they'll go mad
..but surprise infinite lifegain in a white weenie is too much fun to not include...
Re: Tempest world development
by Seravy » 16 Sep 2017, 19:47
Deckbuilding done, not counting the above which I'm waiting on a response for, there are 12 more decks to test.
Out of those,
Requires the AI to learn to bounce beneficial comes into play so I'm moving it to the "currently unplayable" folder.
The Dracoplasm , Reins of Power deck is definitely unplayable, I don't even need to test to guess that. The AI needs to learn to dig for both cards (using Intuition and Scroll Rack) and use them together to steal and sac all enemy creatures. Failing that, Goblin Bombardment can be used to sac them for damage, and Intuition can put Shard Phoenix into the grave.
For the Static Orb deck, the AI will need to learn to use Mind Games with buyback to tap the orb at end of enemy turn to gain an untap phase - and it also needs to learn not to waste the card on casting without buyback. Again, haven't tested, I can't imagine the AI to already know this
Aside from these, I expect the en-Kor deck and white weenie to fail unless the AI already knows how to use their redirection ability, and I'm not sure if Necrologia was done or still in progress for those two decks.
In the stompy, Elven Rite is questionable, I vaguely remember the AI being unable to play the card but I'll need to test that.
Then there is Awakening + Tradewind Rider, the AI already failed using the latter correctly.
I think the rest will work but can't be sure
...but we'll find that out tomorrow.
Out of those,
- | Open
Requires the AI to learn to bounce beneficial comes into play so I'm moving it to the "currently unplayable" folder.
- | Open
The Dracoplasm , Reins of Power deck is definitely unplayable, I don't even need to test to guess that. The AI needs to learn to dig for both cards (using Intuition and Scroll Rack) and use them together to steal and sac all enemy creatures. Failing that, Goblin Bombardment can be used to sac them for damage, and Intuition can put Shard Phoenix into the grave.
For the Static Orb deck, the AI will need to learn to use Mind Games with buyback to tap the orb at end of enemy turn to gain an untap phase - and it also needs to learn not to waste the card on casting without buyback. Again, haven't tested, I can't imagine the AI to already know this
Aside from these, I expect the en-Kor deck and white weenie to fail unless the AI already knows how to use their redirection ability, and I'm not sure if Necrologia was done or still in progress for those two decks.
In the stompy, Elven Rite is questionable, I vaguely remember the AI being unable to play the card but I'll need to test that.
Then there is Awakening + Tradewind Rider, the AI already failed using the latter correctly.
I think the rest will work but can't be sure
...but we'll find that out tomorrow.
Re: Tempest world development
by Seravy » 17 Sep 2017, 07:47
Deck 1
Bigger green creatures with rapid mana acceleration.
AI ignores mana already in its mana pool when casting spells. Even though it had GG from Eladamri's Vineyard, it tapped 2 forests to cast Primal Rage instead of playing a 2GG cost card from hand. Next turn with 3 untapped forests and 2 GG in pool it didn't cast any cards even though it had three creature cards with CMC 4.
Surprisingly, next turn it did use the mana to play an Overgrowth, but only because it could have played it from only the lands.
Overgrowth - The AI doesn't recognize these as producing mana - they don't enable the AI to cast bigger spells. The AI also taps the land it enchants to pay for it when it doesn't have to. (I suggest targeting the newest land in play as it usually taps the oldest first)
This deck is unplayable because it relies on thee two cards heavily.
Next :
Constant Mists with land advantage cards and shadow creatures.
This deck works fair well, except for one very important thing. The AI needs to learn to save mana for casting Constant Mists or Spike Weaver activations when it's about to die - if it lets the damage through, it loses. Semi-unplayable as is.
next : Stompy
Nice, the AI played Elven Rite! That means this one is fully AI playable. (after I remove Tempting Licit, it isn't very good anyway)
4 wins, 3 losses, this deck is fairly good, probably worth setting as medium.
next :
Big creatures played through Dream Halls and Oath of Druids.
The AI is not playing cards through Dream Halls as far as I see and it causes a slowdown when in play.
Based on games played, this will be at least Hard once fixed, but for now it's unplayable due to that.
Next : black control with Necrologia
Checked the thread and it seems you haven't finished that card yet so putting this and the B/W version into unplayable.
That leaves the two decks with en-Kor creatures.
...nope, the AI doesn't consider the ability when declaring attackers/blockers and doesn't use it when they could save a creature. These are also unplayable.
Out of previously unplayable decks, I believe the fittest/recurring one, and the blue control with the draw spells should now work, so retesting those. Silvers should also be fine now with Mox Diamonds working.
Bigger green creatures with rapid mana acceleration.
- | Open
AI ignores mana already in its mana pool when casting spells. Even though it had GG from Eladamri's Vineyard, it tapped 2 forests to cast Primal Rage instead of playing a 2GG cost card from hand. Next turn with 3 untapped forests and 2 GG in pool it didn't cast any cards even though it had three creature cards with CMC 4.
Surprisingly, next turn it did use the mana to play an Overgrowth, but only because it could have played it from only the lands.
Overgrowth - The AI doesn't recognize these as producing mana - they don't enable the AI to cast bigger spells. The AI also taps the land it enchants to pay for it when it doesn't have to. (I suggest targeting the newest land in play as it usually taps the oldest first)
This deck is unplayable because it relies on thee two cards heavily.
Next :
- | Open
Constant Mists with land advantage cards and shadow creatures.
This deck works fair well, except for one very important thing. The AI needs to learn to save mana for casting Constant Mists or Spike Weaver activations when it's about to die - if it lets the damage through, it loses. Semi-unplayable as is.
next : Stompy
- | Open
Nice, the AI played Elven Rite! That means this one is fully AI playable. (after I remove Tempting Licit, it isn't very good anyway)
4 wins, 3 losses, this deck is fairly good, probably worth setting as medium.
next :
- | Open
Big creatures played through Dream Halls and Oath of Druids.
The AI is not playing cards through Dream Halls as far as I see and it causes a slowdown when in play.
Based on games played, this will be at least Hard once fixed, but for now it's unplayable due to that.
Next : black control with Necrologia
- | Open
Checked the thread and it seems you haven't finished that card yet so putting this and the B/W version into unplayable.
That leaves the two decks with en-Kor creatures.
...nope, the AI doesn't consider the ability when declaring attackers/blockers and doesn't use it when they could save a creature. These are also unplayable.
Out of previously unplayable decks, I believe the fittest/recurring one, and the blue control with the draw spells should now work, so retesting those. Silvers should also be fine now with Mox Diamonds working.
Re: Tempest world development
by Seravy » 17 Sep 2017, 09:48
Fittest deck :
Hermit Druid - The AI shouldn't use this until end of main2 if it already has a land in hand. Doing so can prevent it from playing cards and lose the game, especially with multiple copies in play. (AI was holding Spike Feeder and a bunch of 7 CMC creatures and could have played some but instead was digging for lands until it lost)
There was one case when it correctly didn't use it to play a 4 CMC card from 4 lands, but in all other cases it blocked the AI from playing the Spike Feeder and eventually, the bigger cards.
This isn't bad enough to consider the card unplayable but definitely should be improved.
Spike Weaver - no idea what controls this but the AI could have prevented 4 damage and did not. Next turn it also didn't prevent 7 damage. Both cases it had untapped lands and fittest and recurring in play.
The AI also does the mistake of tapping out then losing when I attack even though it has this in play.
Overall, the deck works and is good enough to be retested for higher difficulty, but these are fairly massive problems that should be solved eventually.
Slivers... o_O turn 3 Sliver Queen, the mox sure makes a lot of difference.
Clot Sliver - The AI blocks as though it could regenerate all creatures even if it can only afford one of them.
4 wins 1 loss, I guess the mana is still a problem for this deck. I'll try it again with some nonbasic lands - tempest block nonbasic lands suck but they might still help here.
All the tweaking and the deck still loses most of the time. I have to face reality - old Slivers aren't that good. Every once in a blue moon the deck gets a great draw and wins but usually it's easy level. I'm going to put it at medium anyway - when it gets the good draw it plays as good as a hard or very hard deck.
next : easy version of the blue control.
...to be continued
Hermit Druid - The AI shouldn't use this until end of main2 if it already has a land in hand. Doing so can prevent it from playing cards and lose the game, especially with multiple copies in play. (AI was holding Spike Feeder and a bunch of 7 CMC creatures and could have played some but instead was digging for lands until it lost)
There was one case when it correctly didn't use it to play a 4 CMC card from 4 lands, but in all other cases it blocked the AI from playing the Spike Feeder and eventually, the bigger cards.
This isn't bad enough to consider the card unplayable but definitely should be improved.
Spike Weaver - no idea what controls this but the AI could have prevented 4 damage and did not. Next turn it also didn't prevent 7 damage. Both cases it had untapped lands and fittest and recurring in play.
The AI also does the mistake of tapping out then losing when I attack even though it has this in play.
Overall, the deck works and is good enough to be retested for higher difficulty, but these are fairly massive problems that should be solved eventually.
Slivers... o_O turn 3 Sliver Queen, the mox sure makes a lot of difference.
Clot Sliver - The AI blocks as though it could regenerate all creatures even if it can only afford one of them.
4 wins 1 loss, I guess the mana is still a problem for this deck. I'll try it again with some nonbasic lands - tempest block nonbasic lands suck but they might still help here.
All the tweaking and the deck still loses most of the time. I have to face reality - old Slivers aren't that good. Every once in a blue moon the deck gets a great draw and wins but usually it's easy level. I'm going to put it at medium anyway - when it gets the good draw it plays as good as a hard or very hard deck.
next : easy version of the blue control.
- | Open
...to be continued
Last edited by Seravy on 17 Sep 2017, 10:49, edited 1 time in total.
Re: Tempest world development
by Agetian » 17 Sep 2017, 10:49
I tried to improve Hermit Druid according to your specifications, but haven't checked how well the changes work yet. Basically, the AI won't activate it if it has a mana-producing land in hand, and it'll only activate it in main2 otherwise in order to avoid mana-locking itself. There's no real way to specify "end of main2" because it's a bit too fine-grained prioritization for the current AI model, unfortunately. Let me know if you want it to still activate it even if it has a mana-producing land in hand, that's easy to accomplish.
Spike Weaver - not sure, but I'd need a specific game state (preferably saved via Dump Game State) to recreate conditions for this and see what exactly is preventing the AI from activating it. In my tests, the AI tends to activate it more often than not, so it's a bit difficult to tell why exactly it won't do so in your particular situation and deck. Will definitely look into it once I have a reproducible game state.
- Agetian
Spike Weaver - not sure, but I'd need a specific game state (preferably saved via Dump Game State) to recreate conditions for this and see what exactly is preventing the AI from activating it. In my tests, the AI tends to activate it more often than not, so it's a bit difficult to tell why exactly it won't do so in your particular situation and deck. Will definitely look into it once I have a reproducible game state.
- Agetian
- Agetian
- Programmer
- Posts: 3471
- Joined: 14 Mar 2011, 05:58
- Has thanked: 676 times
- Been thanked: 561 times
Re: Tempest world development
by Seravy » 17 Sep 2017, 10:59
Ok, if end of main2 is too hard for the druid, activate it at end of opponent's turn (or in reaction to the druid getting killed) if there was a land in hand. The goal is to always get a land, unless the mana is needed to play spells. So in a way it's the same thing as the whispers - a card draw effect you use if you don't do anything else, the only difference is you do use it first if you can play the land it fetches as that puts you even at mana.
Which made me realize my specification wasn't very exact. It should activate if it has equal or more lands in hand as it has land plays left for the turn or if it has any cards in hand that have a "discard a land" or "discard X lands" cost/effect.
Weaver : try putting a Survival of Fittest in play. I think there was one when it happened. But the AI only had Swamps as untapped lands. It also had a Recurring Nightmare but that is a sorcery effect so unlikely to matter (unless the AI thinks it's instant, tries to use it and when failing passes priority?). The deck has no other instant effects, so hand shouldn't matter.
The greater problem with this card (and Constant Mists) is the AI not leaving untapped lands to activate it when it's needed, even if they will die.
Which made me realize my specification wasn't very exact. It should activate if it has equal or more lands in hand as it has land plays left for the turn or if it has any cards in hand that have a "discard a land" or "discard X lands" cost/effect.
Weaver : try putting a Survival of Fittest in play. I think there was one when it happened. But the AI only had Swamps as untapped lands. It also had a Recurring Nightmare but that is a sorcery effect so unlikely to matter (unless the AI thinks it's instant, tries to use it and when failing passes priority?). The deck has no other instant effects, so hand shouldn't matter.
The greater problem with this card (and Constant Mists) is the AI not leaving untapped lands to activate it when it's needed, even if they will die.
Re: Tempest world development
by Seravy » 17 Sep 2017, 11:09
Bug :
AI activated Killer Whale 8 times in a row during the declare blockers step when it was an attacking creature. Not very relevant as I lost from that attack but still weird. AI was at 3 life so it's almost as if it thought it was being attacked instead of attacking.
Aside from this the deck works well - in fact so well that it won 5 out of 10. This will be medium instead of easy which means the "medium" version should be hard instead most likely.
With this, the first iteration of testing is complete. I'll make a summary in the next post.
AI activated Killer Whale 8 times in a row during the declare blockers step when it was an attacking creature. Not very relevant as I lost from that attack but still weird. AI was at 3 life so it's almost as if it thought it was being attacked instead of attacking.
Aside from this the deck works well - in fact so well that it won 5 out of 10. This will be medium instead of easy which means the "medium" version should be hard instead most likely.
With this, the first iteration of testing is complete. I'll make a summary in the next post.
Re: Tempest world development
by Seravy » 17 Sep 2017, 11:51
Easy decks : 8
Medium or higher (2nd pass of testing) : 12
Blocked by AI problems : 22
^^^ Outdated
The following cards are blocking decks from AI use, with the number of decks it blocks :
NOTE, I'm editing this list to keep track of the latest situation.
Warrior en-Kor (2) - and all other en-Kor creatures. Ability should be used to either redirect to a target immune to that type of damage, or, to distribute it in a way nothing dies, very much like how it is being assigned with bands. Attack and block declarations need to be aware of this ability.
Knight of Dawn (2) - use to protect itself from damage, enemy spells or gain evasion, the latter only if nothing better can be done with the mana or it wins the game that turn.
Flickering Ward (2) - protect from main enemy color, return to hand if would be destroyed or enchanted creature would be destroyed.
There are also a few cards that aren't blocking decks but the AI needs to learn to use better if possible :
Silver Wyvern (ability)
Skyshroud War Beast (do not play if it would be 0/0 - still broken)
Carrionette (activate ability if enemy can't pay 2)
Pandemonium (play before the creatures)
Volrath's Stronghold (return creatures when it is worth doing - not sure about correct trigger, probably "hand has nothing to play next turn"?)
Spike Feeder (remove counters when the creature will die anyway like Dark Banishing or Wrath of God. Remove counters if it fails to kill opposite creature in combat. Move counters to pump other creatures if needed. )
Sudden Impact (2) - Save mana during own turn to play as early as possible if enemy still has a relevant amount of cards.
Mogg Maniac (1) - The AI should find opportunities when this can block a high power creature to deal large damage to the opponent, and attack aggressively into potential block with those creatures as well. (don't know how to do this)
Fireslinger (3) - AI need to learn it can tap two of them to kill a /2 creature. (Too hard to do, skipping for now)
Coffin Queen (4) - In general it should not be used during upkeep - probably only during enemy turn unless targeting a creature with Haste, unless the AI learns to evaluate which is higher value play, the revived creature or other options in hand. (Probably too much work for too little benefit, not worth delaying the completion of the world for this one.)
Spike ability (1) - not sure how to use this correctly without damage on the stack rules to be honest.
Mind Games (1) - AI needs to learn to hold the card and tap (own) Static Orb to get an untap phase using buyback. Can also work with Winter Orb. (I suspect this deck wouldn't be all that effective to be worth the hard work right now.)
Dracoplasm, Reins of Power, Goblin Bombardment (1) - probably too much work for too little, especially as it has other unplayable cards as well - it's a fun combo deck but the AI would win like 1 out of 100 even if it knew how to play it.
Overgrowth (1) - AI and autopayment needs to recognize "produces an additional" effects. Important but I need to wait for Hanmac who is planning to implement the necessary functions for the TriggerHandler.
Eladamri's Vineyard (1) - AI needs to learn spending main phase 1 mana in its mana pool. Observed this problem with Mana Drain as well, AI never spends that mana either. (required by same deck as Overgrowth which is on hold)
Medium or higher (2nd pass of testing) : 12
Blocked by AI problems : 22
^^^ Outdated
The following cards are blocking decks from AI use, with the number of decks it blocks :
NOTE, I'm editing this list to keep track of the latest situation.
Warrior en-Kor (2) - and all other en-Kor creatures. Ability should be used to either redirect to a target immune to that type of damage, or, to distribute it in a way nothing dies, very much like how it is being assigned with bands. Attack and block declarations need to be aware of this ability.
Knight of Dawn (2) - use to protect itself from damage, enemy spells or gain evasion, the latter only if nothing better can be done with the mana or it wins the game that turn.
Flickering Ward (2) - protect from main enemy color, return to hand if would be destroyed or enchanted creature would be destroyed.
There are also a few cards that aren't blocking decks but the AI needs to learn to use better if possible :
Silver Wyvern (ability)
Skyshroud War Beast (do not play if it would be 0/0 - still broken)
Carrionette (activate ability if enemy can't pay 2)
Pandemonium (play before the creatures)
Volrath's Stronghold (return creatures when it is worth doing - not sure about correct trigger, probably "hand has nothing to play next turn"?)
Spike Feeder (remove counters when the creature will die anyway like Dark Banishing or Wrath of God. Remove counters if it fails to kill opposite creature in combat. Move counters to pump other creatures if needed. )
Sudden Impact (2) - Save mana during own turn to play as early as possible if enemy still has a relevant amount of cards.
Mogg Maniac (1) - The AI should find opportunities when this can block a high power creature to deal large damage to the opponent, and attack aggressively into potential block with those creatures as well. (don't know how to do this)
Last edited by Seravy on 18 Feb 2018, 23:22, edited 25 times in total.
Re: Tempest world development
by Agetian » 17 Sep 2017, 12:44
Ok, I'll see what I can do with these as time permits
Necrologia - yeah, I tried it a couple times but failed, the AI is very unwilling to properly pay its cost for some weird reason. Currently low priority compared to everything else unless I suddenly have an insight as to why exactly it's failing.
- Agetian
Necrologia - yeah, I tried it a couple times but failed, the AI is very unwilling to properly pay its cost for some weird reason. Currently low priority compared to everything else unless I suddenly have an insight as to why exactly it's failing.
- Agetian
- Agetian
- Programmer
- Posts: 3471
- Joined: 14 Mar 2011, 05:58
- Has thanked: 676 times
- Been thanked: 561 times
Re: Tempest world development
by Agetian » 17 Sep 2017, 12:45
That Killer Whale overactivation should already be fixed in a relatively recent build btw, I was looking at a similar bug with Menace and Haste earlier today.
- Agetian
- Agetian
- Agetian
- Programmer
- Posts: 3471
- Joined: 14 Mar 2011, 05:58
- Has thanked: 676 times
- Been thanked: 561 times
Re: Tempest world development
by Agetian » 17 Sep 2017, 12:48
Oh, btw, re: mana payment and autopayment issues - I really wish someone else could take a look at those as well, there are many, many things that could use improvement (detection of spells that produce additional mana in the pool, proper detection of all payment possibilities when having multiple nonbasic lands that produce different colors of mana, etc.). Tried looking into it but seemed way over my head for now 
- Agetian
- Agetian
- Agetian
- Programmer
- Posts: 3471
- Joined: 14 Mar 2011, 05:58
- Has thanked: 676 times
- Been thanked: 561 times
Re: Tempest world development
by Agetian » 17 Sep 2017, 13:01
Ok, tested conditions for Spike Weaver: it looks like the current AI is set to activate it when its life is in danger (which is usually when it's at 5 or lower life). What are the conditions you'd suggest for it?
- Agetian
- Agetian
- Agetian
- Programmer
- Posts: 3471
- Joined: 14 Mar 2011, 05:58
- Has thanked: 676 times
- Been thanked: 561 times
Re: Tempest world development
by Seravy » 17 Sep 2017, 13:17
Always activate ifAgetian wrote:Ok, tested conditions for Spike Weaver: it looks like the current AI is set to activate it when its life is in danger (which is usually when it's at 5 or lower life). What are the conditions you'd suggest for it?
- Agetian
-Life is "in danger" I'd suggest more than 5 for that though, 5 damage can be dealt too easily with spells or evasion creatures so once the AI is there it lost 90% of the time
or
-Incoming damage is at least one quarter of the AI's health
or
-Incoming damage is 5 or more.
or
-Nasty effects trigger when an unblocked attacker hits (Sword of Fire and Ice and the like)
If predicted unblocked enemy creatures would satisfy any of these conditions above,
-leave at least 1 mana untapped (or 2 in case of Constant Mists)
-Do not attack with the weaver if it can be killed by blockers.
(in general, I think 5 is too low for "danger", but that's a different story.)
Edit : In fact it's almost never a good idea to attack into a 3/3 or larger blocker with a Spike Weaver. Keeping it home and transferring counters to other attackers is better than trading it with an enemy 3/3, and the ability is usually more valuable as well.
Re: Tempest world development
by Seravy » 17 Sep 2017, 14:54
Another AI bug : AI didn't play Evacuation when I was attacking with leathal damage. (also, Evacuation will need the general bounce considerations included somehow - the deck can steal creatures and has enters play effects.)
Marked the deck unplayable for the time being.
Marked the deck unplayable for the time being.
75 posts
• Page 3 of 5 • 1, 2, 3, 4, 5
Who is online
Users browsing this forum: No registered users and 73 guests