information moved to wiki

This commit is contained in:
Bug Hunter
2021-03-08 15:10:22 +00:00
parent 882ab0c24b
commit cdca4c08de

View File

@@ -1,28 +0,0 @@
High-level TODOs for AI simulation code:
- When copying the game and simulating, copy what's on the stack. This will
allow AI to use effects that e.g. sacrifice their creatures that are targeted
by removal as well as play counterspells using the simulation framework.
- When evaluating permanents with ETB effects (e.g. oblivion ring), simulate
all possible targets, just like SA targeting during simulation.
- When evaluating what to play during combat, simulate until end of combat, so
that the AI takes into account how much life or chump blockers are saved when
killing an attacking creature, for example.
- When evaluating game state, creature evaluations should take into account the
current turn and phase and what creatures the opponent has - so that for e.g.
if the opponent can kill you next turn with enough attacking creatures, then
effects that produce blockers are rated higher.
- AI is too eager to play spells. For example, the AI may want to play an instant
during draw phase, since that's the best spell to play at that point, whereas
in reality, waiting till MAIN1 and playing a creature or sorcery may be the
better decision.
- AI needs to ration mana somehow. Right now, it's a "greedy play best spell"
system. It fails to take into account situations where playing two cheap
spells is a better alternative than playing one expensive spell whose effect
is better than each cheap spell, but not than both of them.