1
0
mirror of https://github.com/vcmi/vcmi.git synced 2024-11-28 08:48:48 +02:00

BattleAI: divide only positive part of score onto turns count

This commit is contained in:
Andrii Danylchenko 2023-10-03 08:16:12 +03:00
parent 3880ea58b9
commit 9eb9404f28
4 changed files with 187 additions and 55 deletions

View File

@ -190,6 +190,14 @@ BattleAction BattleEvaluator::selectStackAction(const CStack * stack)
if(stack->waited())
{
logAi->debug(
"Moving %s towards hex %s[%d], score: %2f/%2f",
stack->getDescription(),
moveTarget.cachedAttack->attack.defender->getDescription(),
moveTarget.cachedAttack->attack.defender->getPosition().hex,
moveTarget.score,
moveTarget.scorePerTurn);
return goTowardsNearest(stack, moveTarget.positions);
}
else
@ -572,7 +580,7 @@ bool BattleEvaluator::attemptCastingSpell(const CStack * activeStack)
}
else
{
ps.value = scoreEvaluator.calculateExchange(*cachedAttack, *targets, innerCache, state);
ps.value = scoreEvaluator.evaluateExchange(*cachedAttack, *targets, innerCache, state);
}
for(auto unit : allUnits)

View File

@ -18,10 +18,8 @@ AttackerValue::AttackerValue()
}
MoveTarget::MoveTarget()
: positions(), cachedAttack()
: positions(), cachedAttack(), score(EvaluationResult::INEFFECTIVE_SCORE), scorePerTurn(EvaluationResult::INEFFECTIVE_SCORE)
{
score = EvaluationResult::INEFFECTIVE_SCORE;
scorePerTurn = EvaluationResult::INEFFECTIVE_SCORE;
turnsToRich = 1;
}
@ -58,7 +56,7 @@ float BattleExchangeVariant::trackAttack(
auto attackerDamageReduce = AttackPossibility::calculateDamageReduce(defender.get(), unitToUpdate.get(), retaliationDamage, damageCache, hb);
attackValue -= attackerDamageReduce;
dpsScore -= attackerDamageReduce * negativeEffectMultiplier;
dpsScore.ourDamageReduce += attackerDamageReduce;
attackerValue[unitToUpdate->unitId()].isRetalitated = true;
unitToUpdate->damage(retaliationDamage);
@ -80,7 +78,7 @@ float BattleExchangeVariant::trackAttack(
auto collateralDamageReduce = AttackPossibility::calculateDamageReduce(attacker.get(), unitToUpdate.get(), collateralDamage, damageCache, hb);
attackValue -= collateralDamageReduce;
dpsScore -= collateralDamageReduce * negativeEffectMultiplier;
dpsScore.ourDamageReduce += collateralDamageReduce;
unitToUpdate->damage(collateralDamage);
@ -101,7 +99,7 @@ float BattleExchangeVariant::trackAttack(
float defenderDamageReduce = AttackPossibility::calculateDamageReduce(attacker.get(), unitToUpdate.get(), attackDamage, damageCache, hb);
attackValue += defenderDamageReduce;
dpsScore += defenderDamageReduce * positiveEffectMultiplier;
dpsScore.enemyDamageReduce += defenderDamageReduce;
attackerValue[attacker->unitId()].value += defenderDamageReduce;
unitToUpdate->damage(attackDamage);
@ -119,7 +117,7 @@ float BattleExchangeVariant::trackAttack(
}
attackValue += ap.shootersBlockedDmg;
dpsScore += ap.shootersBlockedDmg * positiveEffectMultiplier;
dpsScore.enemyDamageReduce += ap.shootersBlockedDmg;
attacker->afterAttack(ap.attack.shooting, false);
return attackValue;
@ -156,11 +154,11 @@ float BattleExchangeVariant::trackAttack(
if(isOurAttack)
{
dpsScore += defenderDamageReduce * positiveEffectMultiplier;
dpsScore.enemyDamageReduce += defenderDamageReduce;
attackerValue[attacker->unitId()].value += defenderDamageReduce;
}
else
dpsScore -= defenderDamageReduce * negativeEffectMultiplier;
dpsScore.ourDamageReduce += defenderDamageReduce;
defender->damage(attackDamage);
attacker->afterAttack(shooting, false);
@ -182,12 +180,12 @@ float BattleExchangeVariant::trackAttack(
if(isOurAttack)
{
dpsScore -= attackerDamageReduce * negativeEffectMultiplier;
dpsScore.ourDamageReduce += attackerDamageReduce;
attackerValue[attacker->unitId()].isRetalitated = true;
}
else
{
dpsScore += attackerDamageReduce * positiveEffectMultiplier;
dpsScore.enemyDamageReduce += attackerDamageReduce;
attackerValue[defender->unitId()].value += attackerDamageReduce;
}
@ -200,13 +198,18 @@ float BattleExchangeVariant::trackAttack(
#if BATTLE_TRACE_LEVEL>=1
if(!score)
{
logAi->trace("Attack has zero score d:%2f a:%2f", defenderDamageReduce, attackerDamageReduce);
logAi->trace("Attack has zero score def:%2f att:%2f", defenderDamageReduce, attackerDamageReduce);
}
#endif
return score;
}
float BattleExchangeEvaluator::scoreValue(const BattleScore & score) const
{
return score.enemyDamageReduce * getPositiveEffectMultiplier() - score.ourDamageReduce * getNegativeEffectMultiplier();
}
EvaluationResult BattleExchangeEvaluator::findBestTarget(
const battle::Unit * activeStack,
PotentialTargets & targets,
@ -215,7 +218,7 @@ EvaluationResult BattleExchangeEvaluator::findBestTarget(
{
EvaluationResult result(targets.bestAction());
if(!activeStack->waited())
if(!activeStack->waited() && !activeStack->acquireState()->hadMorale)
{
#if BATTLE_TRACE_LEVEL>=1
logAi->trace("Evaluating waited attack for %s", activeStack->getDescription());
@ -230,7 +233,7 @@ EvaluationResult BattleExchangeEvaluator::findBestTarget(
for(auto & ap : targets.possibleAttacks)
{
float score = calculateExchange(ap, targets, damageCache, hbWaited);
float score = evaluateExchange(ap, targets, damageCache, hbWaited);
if(score > result.score)
{
@ -247,9 +250,15 @@ EvaluationResult BattleExchangeEvaluator::findBestTarget(
updateReachabilityMap(hb);
if(result.bestAttack.attack.shooting && hb->battleHasShootingPenalty(activeStack, result.bestAttack.dest))
{
if(!canBeHitThisTurn(result.bestAttack))
return result; // lets wait
}
for(auto & ap : targets.possibleAttacks)
{
float score = calculateExchange(ap, targets, damageCache, hb);
float score = evaluateExchange(ap, targets, damageCache, hb);
if(score >= result.score)
{
@ -269,7 +278,7 @@ MoveTarget BattleExchangeEvaluator::findMoveTowardsUnreachable(
std::shared_ptr<HypotheticBattle> hb)
{
MoveTarget result;
BattleExchangeVariant ev(getPositiveEffectMultiplier(), getNegativeEffectMultiplier());
BattleExchangeVariant ev;
if(targets.unreachableEnemies.empty())
return result;
@ -301,6 +310,9 @@ MoveTarget BattleExchangeEvaluator::findMoveTowardsUnreachable(
auto turnsToRich = (distance - 1) / speed + 1;
auto hexes = closestStack->getSurroundingHexes();
auto enemySpeed = closestStack->speed();
auto speedRatio = speed / static_cast<float>(enemySpeed);
auto penalty = speedRatio > 1 ? 1 : speedRatio;
for(auto hex : hexes)
{
@ -310,13 +322,13 @@ MoveTarget BattleExchangeEvaluator::findMoveTowardsUnreachable(
attack.shootersBlockedDmg = 0; // we do not want to count on it, it is not for sure
auto score = calculateExchange(attack, targets, damageCache, hb);
auto scorePerTurn = score / turnsToRich;
auto score = calculateExchange(attack, turnsToRich, targets, damageCache, hb);
auto scorePerTurn = BattleScore(score.enemyDamageReduce * std::sqrt(penalty / turnsToRich), score.ourDamageReduce);
if(result.scorePerTurn < scorePerTurn)
if(result.scorePerTurn < scoreValue(scorePerTurn))
{
result.scorePerTurn = scorePerTurn;
result.score = score;
result.scorePerTurn = scoreValue(scorePerTurn);
result.score = scoreValue(score);
result.positions = closestStack->getAttackableHexes(activeStack);
result.cachedAttack = attack;
result.turnsToRich = turnsToRich;
@ -357,16 +369,17 @@ std::vector<const battle::Unit *> BattleExchangeEvaluator::getAdjacentUnits(cons
return checkedStacks;
}
std::vector<const battle::Unit *> BattleExchangeEvaluator::getExchangeUnits(
ReachabilityData BattleExchangeEvaluator::getExchangeUnits(
const AttackPossibility & ap,
PotentialTargets & targets,
std::shared_ptr<HypotheticBattle> hb)
{
ReachabilityData result;
auto hexes = ap.attack.defender->getHexes();
if(!ap.attack.shooting) hexes.push_back(ap.from);
std::vector<const battle::Unit *> exchangeUnits;
std::vector<const battle::Unit *> allReachableUnits;
for(auto hex : hexes)
@ -404,7 +417,28 @@ std::vector<const battle::Unit *> BattleExchangeEvaluator::getExchangeUnits(
logAi->trace("Reachability map contains only %d stacks", allReachableUnits.size());
#endif
return exchangeUnits;
return result;
}
for(auto unit : allReachableUnits)
{
auto accessible = !unit->canShoot();
if(!accessible)
{
for(auto hex : unit->getSurroundingHexes())
{
if(ap.attack.defender->coversPos(hex))
{
accessible = true;
}
}
}
if(accessible)
result.melleeAccessible.push_back(unit);
else
result.shooters.push_back(unit);
}
for(int turn = 0; turn < turnOrder.size(); turn++)
@ -412,19 +446,30 @@ std::vector<const battle::Unit *> BattleExchangeEvaluator::getExchangeUnits(
for(auto unit : turnOrder[turn])
{
if(vstd::contains(allReachableUnits, unit))
exchangeUnits.push_back(unit);
result.units.push_back(unit);
}
}
vstd::erase_if(exchangeUnits, [&](const battle::Unit * u) -> bool
vstd::erase_if(result.units, [&](const battle::Unit * u) -> bool
{
return !hb->battleGetUnitByID(u->unitId())->alive();
});
return exchangeUnits;
return result;
}
float BattleExchangeEvaluator::calculateExchange(
float BattleExchangeEvaluator::evaluateExchange(
const AttackPossibility & ap,
PotentialTargets & targets,
DamageCache & damageCache,
std::shared_ptr<HypotheticBattle> hb)
{
BattleScore score = calculateExchange(ap, targets, damageCache, hb);
return scoreValue(score);
}
BattleScore BattleExchangeEvaluator::calculateExchange(
const AttackPossibility & ap,
PotentialTargets & targets,
DamageCache & damageCache,
@ -438,7 +483,7 @@ float BattleExchangeEvaluator::calculateExchange(
&& cb->battleGetGateState() == EGateState::BLOCKED
&& ap.attack.defender->coversPos(BattleHex::GATE_BRIDGE))
{
return EvaluationResult::INEFFECTIVE_SCORE;
return BattleScore(EvaluationResult::INEFFECTIVE_SCORE, 0);
}
std::vector<const battle::Unit *> ourStacks;
@ -447,17 +492,19 @@ float BattleExchangeEvaluator::calculateExchange(
if(hb->battleGetUnitByID(ap.attack.defender->unitId())->alive())
enemyStacks.push_back(ap.attack.defender);
std::vector<const battle::Unit *> exchangeUnits = getExchangeUnits(ap, targets, hb);
vstd::amin(turn, reachabilityMapByTurns.size() - 1);
if(exchangeUnits.empty())
ReachabilityData exchangeUnits = getExchangeUnits(ap, turn, targets, hb);
if(exchangeUnits.units.empty())
{
return 0;
return BattleScore();
}
auto exchangeBattle = std::make_shared<HypotheticBattle>(env.get(), hb);
BattleExchangeVariant v(getPositiveEffectMultiplier(), getNegativeEffectMultiplier());
BattleExchangeVariant v;
for(auto unit : exchangeUnits)
for(auto unit : exchangeUnits.units)
{
if(unit->isTurret())
continue;
@ -481,7 +528,7 @@ float BattleExchangeEvaluator::calculateExchange(
bool canUseAp = true;
for(auto activeUnit : exchangeUnits)
for(auto activeUnit : exchangeUnits.units)
{
bool isOur = exchangeBattle->battleMatchOwner(ap.attack.attacker, activeUnit, true);
battle::Units & attackerQueue = isOur ? ourStacks : enemyStacks;
@ -521,9 +568,16 @@ float BattleExchangeEvaluator::calculateExchange(
return score;
};
if(!oppositeQueue.empty())
auto unitsInOppositeQueueExceptInaccessible = oppositeQueue;
vstd::erase_if(unitsInOppositeQueueExceptInaccessible, [&](const battle::Unit * u)->bool
{
return vstd::contains(exchangeUnits.shooters, u);
});
if(!unitsInOppositeQueueExceptInaccessible.empty())
{
targetUnit = *vstd::maxElementByFun(oppositeQueue, estimateAttack);
targetUnit = *vstd::maxElementByFun(unitsInOppositeQueueExceptInaccessible, estimateAttack);
}
else
{
@ -594,7 +648,7 @@ float BattleExchangeEvaluator::calculateExchange(
v.adjustPositions(melleeAttackers, ap, reachabilityMap);
#if BATTLE_TRACE_LEVEL>=1
logAi->trace("Exchange score: %2f", v.getScore());
logAi->trace("Exchange score: enemy: %2f, our -%2f", v.getScore().enemyDamageReduce, v.getScore().ourDamageReduce);
#endif
return v.getScore();
@ -662,10 +716,26 @@ void BattleExchangeVariant::adjustPositions(
if(notRealizedDamage > ap.attackValue() && notRealizedDamage > attackerValue[ap.attack.attacker->unitId()].value)
{
dpsScore = EvaluationResult::INEFFECTIVE_SCORE;
dpsScore = BattleScore(EvaluationResult::INEFFECTIVE_SCORE, 0);
}
}
bool BattleExchangeEvaluator::canBeHitThisTurn(const AttackPossibility & ap)
{
for(auto pos : ap.attack.attacker->getSurroundingHexes())
{
for(auto u : reachabilityMap[pos])
{
if(u->unitSide() != ap.attack.attacker->unitSide())
{
return true;
}
}
}
return false;
}
void BattleExchangeEvaluator::updateReachabilityMap( std::shared_ptr<HypotheticBattle> hb)
{
const int TURN_DEPTH = 2;

View File

@ -14,6 +14,34 @@
#include "PotentialTargets.h"
#include "StackWithBonuses.h"
struct BattleScore
{
float ourDamageReduce;
float enemyDamageReduce;
BattleScore(float enemyDamageReduce, float ourDamageReduce)
:enemyDamageReduce(enemyDamageReduce), ourDamageReduce(ourDamageReduce)
{
}
BattleScore() : BattleScore(0, 0) {}
float value()
{
return enemyDamageReduce - ourDamageReduce;
}
BattleScore operator+(BattleScore & other)
{
BattleScore result = *this;
result.ourDamageReduce += other.ourDamageReduce;
result.enemyDamageReduce += other.enemyDamageReduce;
return result;
}
};
struct AttackerValue
{
float value;
@ -59,8 +87,8 @@ struct EvaluationResult
class BattleExchangeVariant
{
public:
BattleExchangeVariant(float positiveEffectMultiplier, float negativeEffectMultiplier)
: dpsScore(0), positiveEffectMultiplier(positiveEffectMultiplier), negativeEffectMultiplier(negativeEffectMultiplier) {}
BattleExchangeVariant()
: dpsScore() {}
float trackAttack(
const AttackPossibility & ap,
@ -76,7 +104,7 @@ public:
std::shared_ptr<HypotheticBattle> hb,
bool evaluateOnly = false);
float getScore() const { return dpsScore; }
const BattleScore & getScore() const { return dpsScore; }
void adjustPositions(
std::vector<const battle::Unit *> attackers,
@ -84,12 +112,21 @@ public:
std::map<BattleHex, battle::Units> & reachabilityMap);
private:
float positiveEffectMultiplier;
float negativeEffectMultiplier;
float dpsScore;
BattleScore dpsScore;
std::map<uint32_t, AttackerValue> attackerValue;
};
struct ReachabilityData
{
std::vector<const battle::Unit *> units;
// shooters which are within mellee attack and mellee units
std::vector<const battle::Unit *> melleeAccessible;
// far shooters
std::vector<const battle::Unit *> shooters;
};
class BattleExchangeEvaluator
{
private:
@ -99,12 +136,22 @@ private:
std::vector<battle::Units> turnOrder;
float negativeEffectMultiplier;
float scoreValue(const BattleScore & score) const;
BattleScore calculateExchange(
const AttackPossibility & ap,
PotentialTargets & targets,
DamageCache & damageCache,
std::shared_ptr<HypotheticBattle> hb);
bool canBeHitThisTurn(const AttackPossibility & ap);
public:
BattleExchangeEvaluator(
std::shared_ptr<CBattleInfoCallback> cb,
std::shared_ptr<Environment> env,
float strengthRatio): cb(cb), env(env) {
negativeEffectMultiplier = strengthRatio;
negativeEffectMultiplier = std::sqrt(strengthRatio);
}
EvaluationResult findBestTarget(
@ -113,14 +160,20 @@ public:
DamageCache & damageCache,
std::shared_ptr<HypotheticBattle> hb);
float calculateExchange(
float evaluateExchange(
const AttackPossibility & ap,
PotentialTargets & targets,
DamageCache & damageCache,
std::shared_ptr<HypotheticBattle> hb);
void updateReachabilityMap(std::shared_ptr<HypotheticBattle> hb);
std::vector<const battle::Unit *> getExchangeUnits(const AttackPossibility & ap, PotentialTargets & targets, std::shared_ptr<HypotheticBattle> hb);
ReachabilityData getExchangeUnits(
const AttackPossibility & ap,
uint8_t turn,
PotentialTargets & targets,
std::shared_ptr<HypotheticBattle> hb);
bool checkPositionBlocksOurStacks(HypotheticBattle & hb, const battle::Unit * unit, BattleHex position);
MoveTarget findMoveTowardsUnreachable(
@ -131,6 +184,6 @@ public:
std::vector<const battle::Unit *> getAdjacentUnits(const battle::Unit * unit);
float getPositiveEffectMultiplier() { return 1; }
float getNegativeEffectMultiplier() { return negativeEffectMultiplier; }
float getPositiveEffectMultiplier() const { return 1; }
float getNegativeEffectMultiplier() const { return negativeEffectMultiplier; }
};

View File

@ -274,6 +274,7 @@ void Nullkiller::makeTurn()
bestTask = choseBestTask(bestTasks);
std::string taskDescription = bestTask->toString();
HeroPtr hero = bestTask->getHero();
HeroRole heroRole = HeroRole::MAIN;
@ -292,7 +293,7 @@ void Nullkiller::makeTurn()
logAi->trace(
"Goal %s has low priority %f so decreasing scan depth to gain performance.",
bestTask->toString(),
taskDescription,
bestTask->priority);
}
@ -308,7 +309,7 @@ void Nullkiller::makeTurn()
{
logAi->trace(
"Goal %s has too low priority %f so increasing scan depth to full.",
bestTask->toString(),
taskDescription,
bestTask->priority);
scanDepth = ScanDepth::ALL_FULL;
@ -316,7 +317,7 @@ void Nullkiller::makeTurn()
continue;
}
logAi->trace("Goal %s has too low priority. It is not worth doing it. Ending turn.", bestTask->toString());
logAi->trace("Goal %s has too low priority. It is not worth doing it. Ending turn.", taskDescription);
return;
}
@ -325,7 +326,7 @@ void Nullkiller::makeTurn()
if(i == MAXPASS)
{
logAi->error("Goal %s exceeded maxpass. Terminating AI turn.", bestTask->toString());
logAi->error("Goal %s exceeded maxpass. Terminating AI turn.", taskDescription);
}
}
}